At Capgemini, I designed and developed an Enterprise HR Policy Assistant that provides context-grounded policy answers with source citations using Python, FastAPI, Ollama and RAG.
I engineered its hybrid retrieval pipeline with FAISS semantic search, BM25, Reciprocal Rank Fusion and cross-encoder reranking. I also implemented authentication, role-based document access and user-isolated conversation persistence.
I strengthened the assistant with request quotas, bounded LLM concurrency, metadata-only logging and automated expiry. I established quality checks with 215 regression tests and a sample-policy RAG benchmark, integrating GitHub Actions CI.
In an Authorization Operations Copilot proof of concept, I explored incident triage for simulated payment authorizers using LangGraph and hybrid runbook retrieval. I evaluated agentic and fixed workflows on held-out synthetic cases and validated the prototype with implementation tests.
