experience: 6 yearsfocus: AI Systems & Backend Engineeringstack: Python-firstspecialization:
- LLM pipelines
- Retrieval-Augmented Generation
- Computer Vision
- Time-series ML
- Cloud-native APIsCurrently building production AI systems integrating ML models into scalable backend architectures.
Based in Portugal 🇵🇹
Open to Remote & EU Relocation
LLM-driven conversational system with contextual retrieval.
Architecture:
- Next.js + React + TypeScript frontend
- Gemini (Genkit) inference layer
- RAG pipeline with embedding + retrieval workflow
- Firebase real-time data sync
- Context grounding with citation logic
Project: https://botpresidencial.blog
Computer Vision pipeline for sea turtle identification.
Technical highlights:
- YOLOv8 / YOLOv9 training
- Custom drone dataset curation
- Advanced augmentation strategies
- PyTorch training & evaluation pipeline
- mAP50 ≈ 95%
Applied ML for debt prediction & early anomaly detection.
- Prophet forecasting models
- Feature engineering (lags, trend decomposition)
- Backend risk scoring integration
- Automated monitoring pipeline
Tool to compare Portuguese bank cards, salary accounts and savings products.
Technical highlights:
- Comparison engine for cards, salary accounts & savings products
- AI-powered analysis layer for recommendations
- Compound interest simulator
- Static site deployed via GitHub Pages
ROS pick-and-place system for a Universal Robots UR3: a fixed-pose baseline plus a camera-guided depth point-cloud pipeline, both running identically in simulation or on the real arm.
Technical highlights:
- MoveIt motion planning, Gazebo/RealSense parity between sim and hardware
- RANSAC plane removal + point-cloud clustering for object detection
- ArUco-based camera-to-robot calibration
- ROS Melodic, Python
Messenger-style app where up to 5 AI agents with distinct personalities debate your ideas — each one thinks, speaks, and remembers.
Technical highlights:
- Async streaming from local (Ollama) and cloud (Claude) backends
- Per-agent persona, memory, and temperature settings
- Flet-based desktop UI
- 59-test pytest suite
Real-time desktop app that listens to interview audio and streams AI-generated answers live.
Technical highlights:
- Dual-channel capture: microphone + WASAPI speaker loopback
- Real-time transcription via faster-whisper
- Silence-based question detection with auto-answer pipeline
- Pluggable LLM backends: Ollama (local), Claude, OpenAI
- Desktop UI (pywebview) with streaming answer feed
- LLM orchestration frameworks
- Agent-based systems
- Vector databases & semantic search
- Model serving optimization
- Scalable inference APIs
LinkedIn: https://www.linkedin.com/in/gabriel-est/