AI & Data Science engineering student I build machine learning models, LLM agents and RAG systems, and I care about how they are evaluated: baselines, metrics, error analysis and avoiding data leakage.
- Machine learning & data analysis: classification, regression, time series, ranking, feature engineering, model evaluation
- LLM agents & RAG: multi-agent pipelines (LangGraph, LangChain), tool use, semantic search with embeddings, hybrid retrieval
- Computer vision & signal processing: object detection and tracking (YOLO), acoustic signal classification, inertial sensor data, CNNs
- Experimentation: reproducible pipelines, grouped cross-validation, NDCG / MRR / recall@k, error analysis
| Project | What it does | Stack |
|---|---|---|
| travel-search-agent | LLM agent that turns travel requests into SQL searches, ranks flights with a learned ranker (NDCG@5 0.89 vs 0.41 for price sorting) and gives grounded recommendations | Python, SQL, scikit-learn, LangGraph, OpenAI/Mistral |
| rag-agents-semantic-search | ReAct agent with tools and verified citations over a document base; BM25 vs dense vs hybrid (RRF) retrieval benchmark; FastAPI service | Python, embeddings, BM25, LangGraph, FastAPI |
| email-triage-agents | Multi-agent email triage (intent, priority, sender) with a reviewer agent and human escalation, benchmarked against a TF-IDF baseline with a leakage-free split | Python, LangGraph, scikit-learn, Pydantic |
| pubcam-yolo | Real-time vehicle detection, tracking and counting on public online cameras (YouTube live, HLS, MJPEG, RTSP) with a virtual counting line, CSV export and Docker image | Python, YOLO11, OpenCV, ByteTrack, Docker |
- IRIT (Toulouse, 2026): research intern, multimodal analysis & NLP (TF-IDF, LLM pseudo-labelling, reproducible experiments)
- TAEP (2025โ2026): deep learning instructor for company projects (neural networks, CNN, NLP)
- BandeSoft (2024โ2025): AI intern, LLM agents for test scenario generation and prioritisation
- YowYob Inc (2024): ML intern, AI agents, feature engineering, predictive models

