Enterprise RAG.
Answers grounded in 10,000+ documents. A retrieval system built around relevance, traceability, and continuous evaluation.
Python · LangChain · Pinecone · OpenAI · Next.js · DeepEval · RAGAS
The problem
A large document collection is only useful if an assistant can retrieve the right context and keep answer quality measurable as the system changes.
The engineering
Engineered a production RAG pipeline over 10,000+ documents, combining hybrid retrieval with cross-encoder reranking.
Evaluated answer relevance with RAGAS and automated DeepEval quality scoring in CI.
Included monitoring for p95 latency, token cost, and retrieval-quality drift to make quality and operating behavior visible.
The outcome
Achieved 92% answer relevance in RAGAS evaluation, with automated quality scoring integrated into the delivery workflow.