AI/ML Engineer · M.S. Data Science & Analytics @ Georgia State University
I build production-grade LLM and machine learning systems. Currently a Graduate Research Assistant at Truist Bank, where I developed a RAG evaluation framework for an enterprise AI chatbot using AWS Bedrock and LLM-as-Judge methodology.
Open to Data Scienece & AI/ML Engineer Full Time Roles — based in United States.
- Agentic RAG evaluation framework — production pipeline for measuring retrieval quality, answer faithfulness, and groundedness on an enterprise chatbot. Python, AWS Bedrock, LLM-as-Judge.
- LapLens — full-stack F1 telemetry analytics platform. React + FastAPI, FastF1 data pipeline, deployed on Oracle Cloud + Cloudflare Pages.
- ARIA — 4-stage GenAI pipeline for insurance underwriting. PDF ingestion with custom spaCy NER, XGBoost + SHAP risk classification, FAISS retrieval, and Claude tool-use orchestration. Live demo ↗
- VoxVerity — research study on why audio deepfake detectors that ace benchmarks collapse on real-world audio, with a pre-registered fix and an honestly-reported negative result. PyTorch, wav2vec2, LLM-based explainability.
- Coursework in Deep Learning, Machine Learning, and GenAI Solutions (Spring 2026).
| Project | What it does | Stack |
|---|---|---|
| LapLens | F1 telemetry analytics — lap comparisons, sector analysis, driver performance over a race weekend. | React, FastAPI, FastF1, Oracle Cloud, Cloudflare Pages |
| ARIA · live demo ↗ | GenAI underwriting assistant — PDF → custom NER → XGBoost + SHAP → FAISS retrieval → Claude tool-use synthesis. 4 microservices, 81 tests. | FastAPI, spaCy, XGBoost, SHAP, FAISS, Claude, React, Docker |
| VoxVerity | Audio deepfake detection research — measures the benchmark-to-real-world generalization gap, diagnoses the failure mode via LLM-based explainability, and tests a pre-registered fix. | PyTorch, wav2vec2, FastAPI, Claude, SHAP-style rationale scoring |
| MedSift AI (Hacklytics 2026) | Privacy-first healthcare conversation intelligence — audio in, SOAP notes + risk score + trial matches out. Fully local: Whisper + Presidio PHI redaction + Ollama (LLaMA 3.1), zero cloud APIs. | FastAPI, Whisper, Presidio, Ollama, Streamlit, SQLite FTS5 |
| SparkPath (🥉 AI for Good hackathon) | AI-powered youth career discovery platform — personalized career pathways, mentorship matching, wellness tracking. | React, Node.js, AWS Bedrock, DynamoDB, Socket.io |
Pinned repos below have the full write-ups.
LLMs & GenAI — RAG pipelines, AWS Bedrock, LLM-as-Judge evaluation, prompt engineering, embeddings ML & Deep Learning — PyTorch, TensorFlow, scikit-learn, XGBoost, ensemble methods, SVMs, GAMs Data & Backend — Python (pandas, NumPy), SQL (MySQL, PostgreSQL), FastAPI, R Infra & MLOps — Docker, AWS, GitHub Actions (CI/CD), Oracle Cloud Viz — Tableau, Power BI, matplotlib, Plotly
- LinkedIn — in/tejasvaidya1903
- Email — vaidyatejas02@gmail.com
"Somewhere, something incredible is waiting to be known." — Carl Sagan