Vignesh P A

At Embitel Technologies on Volkswagen Group projects, I build agentic AI systems for hardware-in-the-loop testing and CI/CD failure analysis. My LangGraph-based system triages 25–30 runs each week through enterprise Jenkins and JFrog APIs.

I designed specialized agents to turn 50+ logs, JSON artefacts, PCAP traces, prior incidents, and Confluence documentation into evidence-grounded diagnoses. The system reduced first-pass investigation from roughly eight hours to under one hour while keeping developers in final sign-off.

I also own retrieval work for an internal code and documentation assistant, replacing pure vector search with hybrid BM25 and dense retrieval, Reciprocal Rank Fusion, and cross-encoder reranking. This improved RAGAS answer relevancy from 0.34 to 0.73 on a fixed evaluation set.

Beyond RAG, I fine-tuned Mistral 7B for C++ unit-test generation and built multi-agent research and codebase assistants. I bring an automotive systems background in HIL, CAN, and SOME/IP to domain-heavy AI problems where generic RAG falls short.