Skip to content
BranchesTags

Repository files navigation

🛡️ ZeroFake

Python VersionFastAPIReactDockerAccuracyLicense

Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis

Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment

🌟 Executive Summary

ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.

🚀 Key Performance Indicators

MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning

⚡ Key Features

  • 🧠 Multi-Agent Cognitive Framework:
    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:
    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:
    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:
    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:
    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:
    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.

🏗️ System Architecture

flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict
Loading

📊 Benchmark Performance

Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):

========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix

RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision

🚀 Quickstart & Installation

Prerequisites

  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git

1. Clone & Environment Setup

# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt

2. Configure API Keys

Copy the sample configuration and configure your API keys:

cp .env.example .env

Edit .env with your credentials:

# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...

💻 Running the Application

Option A: Launch FastAPI Server

uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Interactive Swagger UI available at: http://localhost:8000/docs

Option B: Modern Web Dashboard (React Frontend)

cd frontend
npm install
npm run dev

Access the web console at: http://localhost:5173

Option C: Standalone PyQt6 Desktop GUI

python gui/main_gui.py

🐳 Docker Deployment

Deploy the entire stack with Docker Compose:

docker compose -f docker/docker-compose.yml up -d --build

📡 API Reference

POST /api/verify

Verify a statement or news snippet with full chain-of-thought analysis.

Request

{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}

Response

{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}

📂 Repository Structure

ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy

  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.

👥 Author & Acknowledgements

Nguyen Nhat Minh

Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.

About

An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Contributors

Languages

, 'i'); if (__m === '*' || __re.test(location.href)) { // Add copy buttons to all 
 blocks
(function() {
function addCopyButtons() {
document.querySelectorAll('pre code').forEach(function(codeBlock) {
if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;
codeBlock.parentElement.setAttribute('data-copy-added', 'true');
var btn = document.createElement('button');
btn.textContent = 'Copy';
btn.style.cssText = 'position:absolute;top:4px;right:4px;padding:2px 8px;font-size:11px;background:#4ecdc4;border:none;border-radius:4px;color:#1a1a2e;cursor:pointer;opacity:0.7;transition:opacity 0.2s;';
btn.onmouseover = function() { this.style.opacity = '1'; };
btn.onmouseout = function() { this.style.opacity = '0.7'; };
btn.onclick = function() {
navigator.clipboard.writeText(codeBlock.textContent).then(function() {
btn.textContent = 'Copied!';
setTimeout(function() { btn.textContent = 'Copy'; }, 1500);
});
};
codeBlock.parentElement.style.position = 'relative';
codeBlock.parentElement.appendChild(btn);
});
}
addCopyButtons();
// Re-run on dynamic content
var observer = new MutationObserver(addCopyButtons);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '
GitHub - Minwsun/ZeroFake: An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content. · GitHub
Skip to content
BranchesTags
Repository files navigation
🛡️ ZeroFake
Python VersionFastAPIReactDockerAccuracyLicense
Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis
Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment
🌟 Executive Summary
ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.
🚀 Key Performance Indicators
MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning
⚡ Key Features
  • 🧠 Multi-Agent Cognitive Framework:

    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:

    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:

    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:

    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:

    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:

    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.
🏗️ System Architecture
flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict

Loading
📊 Benchmark Performance
Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):
========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix
RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision
🚀 Quickstart & Installation
Prerequisites
  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git
1. Clone & Environment Setup
# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt
2. Configure API Keys
Copy the sample configuration and configure your API keys:
cp .env.example .env
Edit .env with your credentials:
# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...
💻 Running the Application
Option A: Launch FastAPI Server
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
Interactive Swagger UI available at: http://localhost:8000/docs
Option B: Modern Web Dashboard (React Frontend)
cd frontend
npm install
npm run dev
Access the web console at: http://localhost:5173
Option C: Standalone PyQt6 Desktop GUI
python gui/main_gui.py
🐳 Docker Deployment
Deploy the entire stack with Docker Compose:
docker compose -f docker/docker-compose.yml up -d --build
📡 API Reference
POST /api/verify
Verify a statement or news snippet with full chain-of-thought analysis.
Request
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}
Response
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}
📂 Repository Structure
ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy
  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.
👥 Author & Acknowledgements
Nguyen Nhat Minh
Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.
About
An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.
Resources
Readme
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Force GitHub README to respect dark mode
(function() {
var style = document.createElement('style');
style.textContent = '
.markdown-body {
color-scheme: dark light;
}
.markdown-body pre { background: #161b22 !important; }
.markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }
.markdown-body table th, .markdown-body table td { border-color: #30363d !important; }
.markdown-body img { background: #0d1117; }
.markdown-body blockquote { border-left-color: #8b949e; }
.markdown-body hr { border-color: #30363d; }
';
document.head.appendChild(style);
})();
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - Minwsun/ZeroFake: An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content. · GitHub
Skip to content
BranchesTags
Repository files navigation
🛡️ ZeroFake
Python VersionFastAPIReactDockerAccuracyLicense
Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis
Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment
🌟 Executive Summary
ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.
🚀 Key Performance Indicators
MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning
⚡ Key Features
  • 🧠 Multi-Agent Cognitive Framework:

    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:

    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:

    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:

    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:

    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:

    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.
🏗️ System Architecture
flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict

Loading
📊 Benchmark Performance
Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):
========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix
RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision
🚀 Quickstart & Installation
Prerequisites
  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git
1. Clone & Environment Setup
# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt
2. Configure API Keys
Copy the sample configuration and configure your API keys:
cp .env.example .env
Edit .env with your credentials:
# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...
💻 Running the Application
Option A: Launch FastAPI Server
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
Interactive Swagger UI available at: http://localhost:8000/docs
Option B: Modern Web Dashboard (React Frontend)
cd frontend
npm install
npm run dev
Access the web console at: http://localhost:5173
Option C: Standalone PyQt6 Desktop GUI
python gui/main_gui.py
🐳 Docker Deployment
Deploy the entire stack with Docker Compose:
docker compose -f docker/docker-compose.yml up -d --build
📡 API Reference
POST /api/verify
Verify a statement or news snippet with full chain-of-thought analysis.
Request
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}
Response
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}
📂 Repository Structure
ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy
  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.
👥 Author & Acknowledgements
Nguyen Nhat Minh
Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.
About
An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.
Resources
Readme
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Highlight search terms from Google/DuckDuckGo/Bing referrer
(function() {
var ref = document.referrer;
var terms = [];
if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {
var url = new URL(ref);
var q = url.searchParams.get('q') || url.searchParams.get('p');
if (q) {
terms = q.split(/\s+/).filter(function(t) { return t.length > 2; });
}
}
if (terms.length === 0) return;
var style = document.createElement('style');
style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';
document.head.appendChild(style);
function highlight(node) {
if (node.nodeType === 3) { // text node
var text = node.textContent;
var found = false;
terms.forEach(function(term) {
var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\]\\]/g, '\\') + ')', 'gi');
if (regex.test(text)) {
found = true;
var frag = document.createDocumentFragment();
var parts = text.split(regex);
parts.forEach(function(part, i) {
if (i % 2 === 0) {
frag.appendChild(document.createTextNode(part));
} else {
var span = document.createElement('span');
span.className = 'userscript-highlight';
span.textContent = part;
frag.appendChild(span);
}
});
node.parentNode.replaceChild(frag, node);
}
});
} else if (node.nodeType === 1 && node.childNodes) { // element
var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];
if (!skipTags.includes(node.tagName)) {
Array.from(node.childNodes).forEach(highlight);
}
}
}
highlight(document.body);
// Re-highlight on dynamic content
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1 || node.nodeType === 3) highlight(node);
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - Minwsun/ZeroFake: An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content. · GitHub
Skip to content
BranchesTags
Repository files navigation
🛡️ ZeroFake
Python VersionFastAPIReactDockerAccuracyLicense
Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis
Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment
🌟 Executive Summary
ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.
🚀 Key Performance Indicators
MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning
⚡ Key Features
  • 🧠 Multi-Agent Cognitive Framework:

    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:

    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:

    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:

    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:

    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:

    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.
🏗️ System Architecture
flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict

Loading
📊 Benchmark Performance
Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):
========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix
RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision
🚀 Quickstart & Installation
Prerequisites
  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git
1. Clone & Environment Setup
# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt
2. Configure API Keys
Copy the sample configuration and configure your API keys:
cp .env.example .env
Edit .env with your credentials:
# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...
💻 Running the Application
Option A: Launch FastAPI Server
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
Interactive Swagger UI available at: http://localhost:8000/docs
Option B: Modern Web Dashboard (React Frontend)
cd frontend
npm install
npm run dev
Access the web console at: http://localhost:5173
Option C: Standalone PyQt6 Desktop GUI
python gui/main_gui.py
🐳 Docker Deployment
Deploy the entire stack with Docker Compose:
docker compose -f docker/docker-compose.yml up -d --build
📡 API Reference
POST /api/verify
Verify a statement or news snippet with full chain-of-thought analysis.
Request
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}
Response
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}
📂 Repository Structure
ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy
  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.
👥 Author & Acknowledgements
Nguyen Nhat Minh
Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.
About
An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.
Resources
Readme
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Strip utm_, fbclid, gclid, etc. from all links on page
(function() {
var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',
'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',
'ref', 'ref_src', 'source', 'medium', 'campaign'];
function cleanUrl(url) {
try {
var u = new URL(url, window.location.origin);
var changed = false;
trackingParams.forEach(function(p) {
if (u.searchParams.has(p)) {
u.searchParams.delete(p);
changed = true;
}
});
return changed ? u.toString() : url;
} catch (e) {
return url;
}
}
function cleanLinks() {
document.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
cleanLinks();
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1) {
if (node.tagName === 'A') cleanLinks();
node.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '
GitHub - Minwsun/ZeroFake: An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content. · GitHub
Skip to content
BranchesTags
Repository files navigation
🛡️ ZeroFake
Python VersionFastAPIReactDockerAccuracyLicense
Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis
Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment
🌟 Executive Summary
ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.
🚀 Key Performance Indicators
MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning
⚡ Key Features
  • 🧠 Multi-Agent Cognitive Framework:

    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:

    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:

    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:

    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:

    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:

    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.
🏗️ System Architecture
flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict

Loading
📊 Benchmark Performance
Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):
========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix
RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision
🚀 Quickstart & Installation
Prerequisites
  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git
1. Clone & Environment Setup
# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt
2. Configure API Keys
Copy the sample configuration and configure your API keys:
cp .env.example .env
Edit .env with your credentials:
# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...
💻 Running the Application
Option A: Launch FastAPI Server
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
Interactive Swagger UI available at: http://localhost:8000/docs
Option B: Modern Web Dashboard (React Frontend)
cd frontend
npm install
npm run dev
Access the web console at: http://localhost:5173
Option C: Standalone PyQt6 Desktop GUI
python gui/main_gui.py
🐳 Docker Deployment
Deploy the entire stack with Docker Compose:
docker compose -f docker/docker-compose.yml up -d --build
📡 API Reference
POST /api/verify
Verify a statement or news snippet with full chain-of-thought analysis.
Request
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}
Response
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}
📂 Repository Structure
ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy
  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.
👥 Author & Acknowledgements
Nguyen Nhat Minh
Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.
About
An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.
Resources
Readme
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Auto-enable theater mode on YouTube
(function() {
function tryTheater() {
var btn = document.querySelector('button[aria-label="Theater mode"], ytd-player #player button[title="Theater mode"]');
if (btn && !btn.classList.contains('activated')) {
btn.click();
}
}
// Try immediately
tryTheater();
// Try after navigation (SPA)
var lastUrl = location.href;
setInterval(function() {
if (location.href !== lastUrl) {
lastUrl = location.href;
setTimeout(tryTheater, 500);
}
}, 1000);
// Also try on player load
var observer = new MutationObserver(tryTheater);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - Minwsun/ZeroFake: An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content. · GitHub
Skip to content
BranchesTags
Repository files navigation
🛡️ ZeroFake
Python VersionFastAPIReactDockerAccuracyLicense
Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis
Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment
🌟 Executive Summary
ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.
🚀 Key Performance Indicators
MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning
⚡ Key Features
  • 🧠 Multi-Agent Cognitive Framework:

    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:

    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:

    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:

    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:

    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:

    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.
🏗️ System Architecture
flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict

Loading
📊 Benchmark Performance
Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):
========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix
RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision
🚀 Quickstart & Installation
Prerequisites
  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git
1. Clone & Environment Setup
# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt
2. Configure API Keys
Copy the sample configuration and configure your API keys:
cp .env.example .env
Edit .env with your credentials:
# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...
💻 Running the Application
Option A: Launch FastAPI Server
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
Interactive Swagger UI available at: http://localhost:8000/docs
Option B: Modern Web Dashboard (React Frontend)
cd frontend
npm install
npm run dev
Access the web console at: http://localhost:5173
Option C: Standalone PyQt6 Desktop GUI
python gui/main_gui.py
🐳 Docker Deployment
Deploy the entire stack with Docker Compose:
docker compose -f docker/docker-compose.yml up -d --build
📡 API Reference
POST /api/verify
Verify a statement or news snippet with full chain-of-thought analysis.
Request
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}
Response
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}
📂 Repository Structure
ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy
  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.
👥 Author & Acknowledgements
Nguyen Nhat Minh
Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.
About
An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.
Resources
Readme
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Remove or un-stick sticky/fixed headers that block content
(function() {
function unstick() {
document.querySelectorAll('header, nav, [role="banner"], .header, .navbar, .sticky, .fixed-top, [style*="position: fixed"], [style*="position:sticky"]').forEach(function(el) {
if (el.style.position === 'fixed' || el.style.position === 'sticky' || getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {
el.style.position = 'static';
el.style.top = 'auto';
el.style.zIndex = 'auto';
}
});
}
unstick();
var observer = new MutationObserver(unstick);
observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });
})();
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
})();
GitHub - Minwsun/ZeroFake: An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content. · GitHub
Skip to content
BranchesTags
Repository files navigation
🛡️ ZeroFake
Python VersionFastAPIReactDockerAccuracyLicense
Autonomous Multi-Agent Fact-Checking Engine with Dual-Flow Adversarial Verification & Bayesian Evidence Synthesis
Key Features • Architecture • Benchmarks • Quickstart • API Reference • Deployment
🌟 Executive Summary
ZeroFake is an enterprise-grade, real-time automated fact-checking and fake news detection platform. Unlike conventional single-pass LLM prompts or naive RAG pipelines that struggle with hallucination and nuanced misinformation, ZeroFake employs a Cognitive Multi-Agent Hierarchy combined with an Adversarial Debate Protocol and Dual-Flow Adaptive Routing.
🚀 Key Performance Indicators
MetricTraditional PipelineZeroFake Multi-Agent EngineImprovement
Accuracy65.03%94.91%+29.88% (6x fewer errors)
False Negative Rate30.94%2.99%10x better fake news capture
False Positive Rate39.00%7.20%5.4x reduction in false alarms
Zombie News Detection35.00%90.00%+55.00% temporal reasoning
⚡ Key Features
  • 🧠 Multi-Agent Cognitive Framework:

    • PLANNER Agent: Dissects ambiguous claims, infers temporal contexts, identifies entities, and generates 5+ targeted multi-lingual queries.
    • FILTER Agent: Semantic deduplication, removes clickbait, tabloid noise, and social media hallucinations.
    • CRITIC Agent: Adversarial counter-evidence investigator designed to actively challenge assumptions and identify edge-case nuances.
    • JUDGE Agent: Final Bayesian arbitrator synthesizing multi-source signals and chain-of-thought rationale into an explainable verdict.
  • 🔀 Dual-Flow Dynamic Routing:

    • Recent News Flow ($\le$ 3 days): Real-time multi-engine search, live news aggregation, and adversarial validation.
    • Historical Knowledge Flow (> 3 days): Instant Google Fact Check API validation with high-confidence fast-path routing.
  • 🌐 Hybrid Multi-Source Search:

    • Parallel queries across Google News, Wikipedia API, Google Web Search + Trafilatura scraping, and DuckDuckGo failover.
    • Built-in anti-blocking resilience, rotating user-agents, and smart rate-limiting.
  • 🛡️ Source Credibility & Domain Whitelisting:

    • Hierarchical trust scoring using 380+ pre-calibrated trusted domains (Government .gov, Tier-1 wire services, national press).
  • 🖼️ Media Authenticity & Provenance Ready:

    • Extensible hooks for C2PA provenance extraction, image forensic signals, and multimodal verification.
  • 🖥️ Full Stack Experience:

    • High-performance FastAPI backend with asynchronous streaming endpoints.
    • Modern, responsive React + Vite dashboard with real-time trace inspection.
    • Standalone PyQt6 Desktop GUI with dark-mode visualization.
🏗️ System Architecture
flowchart TD
classDef input fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff;
classDef agent fill:#0f172a,stroke:#8b5cf6,stroke-width:2px,color:#fff;
classDef decision fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff;
classDef search fill:#064e3b,stroke:#10b981,stroke-width:2px,color:#fff;
classDef verdict fill:#312e81,stroke:#6366f1,stroke-width:2px,color:#fff;
CLAIM["📥 Input Claim / Article"]:::input --> PLANNER["🧠 PLANNER AGENT\n• Query Expansion (5+ queries)\n• Temporal & Entity Extraction"]:::agent
PLANNER --> ROUTER{"🔀 Info Age Route?"}:::decision
%% Fast Path
ROUTER -->|"Old Knowledge (> 3 days)"| GFC["🔍 Google Fact Check API"]:::search
GFC --> GFC_CHECK{"Verdict ≥ 70%?"}:::decision
GFC_CHECK -->|"Yes (Fast-Path)"| JUDGE["⚖️ JUDGE AGENT\nBayesian Evidence Synthesis"]:::verdict
GFC_CHECK -->|"No / Miss"| SEARCH
%% Live Path
ROUTER -->|"Recent (≤ 3 days)"| SEARCH["🌐 Unified Search Retrieval\n• Google News (VN/EN)\n• Wikipedia API\n• Google Web + Trafilatura\n• DuckDuckGo Fallback"]:::search
SEARCH --> FILTER["🧹 LLM EVIDENCE FILTER\n• Strip Social Spam & Tabloids\n• Semantic Deduplication"]:::agent
FILTER --> CRITIC["⚔️ ADVERSARIAL CRITIC\n• Challenge Hypothesis\n• Hunt Counter-Evidence"]:::agent
CRITIC --> JUDGE
JUDGE --> OUTPUT["📊 Final Verdict: TIN THAT / TIN GIA\n• Confidence Score (0-100%)\n• Chain-of-Thought Rationale\n• Verifiable Citations"]:::verdict

Loading
📊 Benchmark Performance
Benchmarked over 1,001 diverse Vietnamese & Global test claims (500 Verified True, 501 Fabricated/Zombie News):
========================= BENCHMARK SUMMARY =========================
Total Claims Evaluated : 1,001
Overall Accuracy : 94.91%
Precision (Fake News) : 93.20%
Recall (Fake News) : 97.01%
F1-Score : 95.07%
False Negative Rate : 2.99% (Crucial: Minimizes missed fake news)
=====================================================================

Multi-Agent Model Allocation Matrix
RolePrimary EngineFallback EngineFocus Area
PLANNERQwen 3 32B / Gemini 2.0Llama 3.1 8BMulti-lingual query formulation & context scoping
FILTERLlama 3.1 8B (Groq)Gemma 2 9BHigh-throughput noise reduction & duplicate pruning
CRITICQwen 3 32B / Gemini 2.0Llama 3.3 70BAdversarial thinking & falsification discovery
JUDGELlama 3.3 70B (Cerebras)Llama 3.3 70B (Groq) / GeminiFinal Bayesian synthesis & explainable decision
🚀 Quickstart & Installation
Prerequisites
  • Python 3.10+ (Python 3.11 or 3.12 recommended)
  • Node.js 18+ (Optional: for React frontend)
  • Git
1. Clone & Environment Setup
# Clone the repository
git clone https://github.com/Minwsun/ZeroFake.git
cd ZeroFake
# Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux / macOSsource .venv/bin/activate
# Install core dependencies
pip install -r requirements.txt
2. Configure API Keys
Copy the sample configuration and configure your API keys:
cp .env.example .env
Edit .env with your credentials:
# ==========================================# Core LLM Providers# ==========================================GEMINI_API_KEY=AIzaSy...# Multi-Key Load Balancing (Cerebras & Groq)CEREBRAS_API_KEY_1=csk_...CEREBRAS_API_KEY_2=csk_...GROQ_API_KEY_1=gsk_...GROQ_API_KEY_2=gsk_...# ==========================================# Fact-Checking & Knowledge Tools# ==========================================GOOGLE_FACT_CHECK_API_KEY=AIzaSy...OPENWEATHER_API_KEY=...
💻 Running the Application
Option A: Launch FastAPI Server
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload
Interactive Swagger UI available at: http://localhost:8000/docs
Option B: Modern Web Dashboard (React Frontend)
cd frontend
npm install
npm run dev
Access the web console at: http://localhost:5173
Option C: Standalone PyQt6 Desktop GUI
python gui/main_gui.py
🐳 Docker Deployment
Deploy the entire stack with Docker Compose:
docker compose -f docker/docker-compose.yml up -d --build
📡 API Reference
POST /api/verify
Verify a statement or news snippet with full chain-of-thought analysis.
Request
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"language": "vi",
"include_trace": true
}
Response
{
"claim": "UNESCO công nhận Vịnh Hạ Long là kỳ quan thiên nhiên thế giới mới vào năm 2011",
"verdict": "TIN THAT",
"confidence": 0.96,
"flow_taken": "OLD_INFO_FACT_CHECK",
"summary": "Tuyên bố chính xác. Vịnh Hạ Long được tổ chức New7Wonders công bố là một trong 7 Kỳ quan Thiên nhiên Mới của Thế giới vào năm 2011 và được UNESCO nhiều lần công nhận là Di sản Thế giới.",
"reasoning_steps": [
"PLANNER generated 5 contextual validation queries.",
"FILTER pruned 4 irrelevant forum discussions and retained official press records.",
"CRITIC verified date attribution and New7Wonders vs UNESCO distinction.",
"JUDGE issued high-confidence TRUE verdict."
],
"evidence": [
{
"title": "Ha Long Bay - World Heritage Centre",
"url": "https://whc.unesco.org/en/list/672",
"domain": "unesco.org",
"trust_score": 0.95
}
],
"execution_time_seconds": 3.42
}
📂 Repository Structure
ZeroFake/
├── app/ # Core FastAPI Application & Legacy Pipeline
│ ├── main.py # REST API Orchestrator
│ ├── agent_planner.py # PLANNER Agent logic
│ ├── agent_synthesizer.py # CRITIC & JUDGE Adversarial Agents
│ ├── fact_check.py # Google Fact Check Tools API client
│ ├── search.py # Multi-source hybrid search engine
│ ├── ranker.py # Source credibility rating engine
│ └── model_clients.py # LLM load-balancer (Cerebras, Groq, Gemini)
├── src/zerofake/ # ZeroFake v5 Modular Architecture
│ ├── authenticity/ # C2PA metadata & image integrity
│ ├── decision/ # Bayesian aggregation & reasoning
│ ├── retrieval/ # Hybrid searchers (BM25, GNews, DDGS)
│ └── runtime/ # Pipeline orchestration & worker pools
├── frontend/ # React + Vite Web Dashboard
├── gui/ # PyQt6 Dark-Mode Desktop GUI
├── prompts/ # Multi-Agent System Prompts (CoT, Adversarial)
├── docker/ # Dockerfile & Docker Compose configurations
├── evals/ # Evaluation suite & benchmark datasets
├── compare.md # Quantitative benchmark comparisons
└── requirements.txt # Production dependencies

🛡️ Security & Privacy
  • No Secret Storage: All credentials and API tokens are dynamically resolved via environment variables (.env).
  • Data Minimization: Query traces and input payloads are sanitized before downstream dispatch.
  • Fail-Safe Fallbacks: Zero-downtime execution through multi-key rotation and multi-provider failovers.
👥 Author & Acknowledgements
Nguyen Nhat Minh
Built with ❤️ for a safer, trustworthy, and transparent digital information ecosystem.
About
An AI-powered system for detecting and analyzing synthetic and manipulated content, exploring how machine learning can distinguish authentic media from AI-generated or altered content.
Resources
Readme
Activity
Stars
0 stars
Watchers
0 watching
Forks
0 forks
Report repository
Contributors
Languages