- 🤖 Agentic AI and RAG systems in production at Grapeworks AI: multi-agent orchestration, GraphRAG pipelines, and LLM tool-calling shipped for live clients.
- 🏆 Top 5 worldwide out of thousands of teams in a Kaggle LLM agentic-retrieval competition, built on a 12-stage hybrid retrieval pipeline with multi-LLM reranking over a multilingual legal corpus.
- 📈 Shipped a RAG document-retrieval system serving 200+ daily users, lifting match accuracy roughly 30% with embedding-based retrieval and vector search.
- 🎓 MSc Data Science, University of Glasgow (Russell Group), expected Distinction.
Building multi-agent automation and GraphRAG products at Grapeworks AI, going deeper on LangGraph, LLM fine-tuning (LoRA), retrieval and reranking, and distributed training.
AI / LLM
Build & Ship
Data & Cloud
Cross-lingual Swiss legal citation retrieval, ranked in the top 5 of thousands of teams. A 12-stage hybrid pipeline fusing BM25, dense embeddings, and graph co-citation, topped with a cross-encoder reranker and a multi-LLM agentic voting layer.
Python · PyTorch · FAISS · BGE-M3 · Qwen2.5 · DeepSeek
💹 RAG for Complex Financial Data
Retrieval-Augmented Generation over complex, multi-source financial documents, with recursive validation loops that keep numerical KPI extraction accurate.
Python · LangChain · Vector Search · Embeddings
🔗 More work, including production systems built at Grapeworks AI, on my portfolio.
