Bhavneet Bhoria

Building and open-sourcing eval-first RAG and agentic systems alongside my MSc, each designed around regression-tested retrieval quality rather than asserted outputs. • DocRAG (agentic RAG on LangGraph, hybrid BM25 + ChromaDB): diagnosed context precision at 0.49 and lifted it to 0.78 with cross-encoder reranking, measuring the recall tradeoff (0.85) instead of hiding it. Scored on the live pipeline against a golden set with bootstrap 95% confidence intervals. • Red-teamed the same system for indirect pr

Machine Learning Engineer at a B2B AI/ML startup delivering predictive analytics and NLP automation to fintech and logistics clients. • Designed and deployed a document-retrieval pipeline from scratch (early production RAG) that cut internal triage time by ~40%, directly informing the team's LLM adoption roadmap. • Achieved a 12% accuracy gain over baseline on domain-specific NLP classification via transfer learning across 50k+ labelled samples. • Served fine-tuned models via FastAPI endpoints container