At Cyient, I built a LangGraph requirement-to-test agent for Powertrain and After-Treatment teams. It retrieves evidence from DOORS requirements and prior TPT suites, drafts test cases, and routes low-confidence results through human review.
I added Langfuse evaluation and an eval harness to track requirement coverage, hallucinations, and reviewer acceptance. I also shipped a GenAI copilot for SIL failure triage, grounding citation-only answers in TPT logs, Polyspace findings, and past RCA notes.
At Robert Bosch Engineering and Business Solutions, I trained XGBoost and Scikit-learn models to score pre-regression failure risk from historical TPT results and coverage signals. I also built a classification pipeline to flag DOORS requirements with thin test coverage and published model results in Power BI and Tableau.
