← All projects
Nexus: RAG Evaluation & Knowledge System
Category :
GenAI & LLMs
Dates :
Aug 2025 - Present
Overview
A retrieval system over 10,000+ documents, tuned and measured like a real product.
What I built
Reached 95%+ retrieval relevance on a fixed regression query set by tuning hybrid search, chunking, metadata filters, and re-ranking.
Measured changes with precision@k and recall@k; grounded answers in retrieved passages and rejected malformed output through schema validation.
Built ingestion for PDFs, Markdown, and scraped web content with chunking, metadata enrichment, tagging, and NER before indexing.
Tech stack
- Python
- RAG
- Embeddings
- Hybrid Search
- Re-Ranking
- NER
- Schema Validation
- Evaluation Harnesses
Key results
95%+ retrieval relevance
10,000+ documents
