Nikhil Mourya

I’m an ML engineer currently building production-grade LLM applications—especially multi-agent orchestration and RAG pipelines that are measurable, testable, and fast at scale. I shipped systems that process 10K+ requests/day, combining strong retrieval with grounded generation and practical deployment patterns.

Most recently, I built an India-specific resume scoring system using transformer embeddings and FAISS vector search, then fine-tuned on a 75K-resume dataset. My RAG retrieval pipeline (LangChain/LangGraph + OpenAI API) improved shortlist precision by 35% versus a keyword-matching baseline at 10K+ resumes/day, and my Docker + AWS MLOps deployment served traffic through a REST API, increasing platform engagement by 40% by integrating into recruiter workflows.

I also build tools and research prototypes that feel production-ready. I created HiveMind, a 5-agent autonomous deep research system that loops until a deterministic confidence threshold is met to prevent hallucinated certainty, and I shipped CodeLens, an offline VS Code extension for semantic code search with AST-based chunking and sub-10ms ANN latency. Alongside applied engineering, I’ve validated parameter-efficient fine-tuning with PEGASUS + LoRA (dramatically reducing trainable parameters) and contributed to IIT Kharagpur research on structured pruning for biomedical segmentation (IoU > 0.95).