Badrish M Srinivasan

Built a multi-agent research assistant in Python (LangGraph, MCP) that grounds answers in cited sources and surfaces evidence gaps in a Streamlit interface. Designed a 400-case citation evaluation set and scored outputs against baselines with Hugging Face Transformers and LangSmith to separate real prompt gains from noise. Maintained 30+ evaluation and guardrail checks in CI, cutting manual citation review time by 70% and visible citation errors by 40%.

Contributed to an enterprise RAG assistant (FastAPI, Azure OpenAI) that linked device faults to 250+ field manuals and 10,000+ incident records in PostgreSQL. Deployed containerised applications on AWS with Docker, Kubernetes, and Helm, and ran Celery and Amazon SQS background jobs. Tuned retrieval with firmware engineers using CloudWatch, Grafana, and OpenTelemetry during live incident response.

Cut forecast error (MAPE) from 22% to 14% for operations workload planning. Built forecasting models in Python with pandas, scikit-learn, and SQL, then validated them with systematic checks. Shipped MLOps modules (Azure Machine Learning, GitHub Actions, Docker) with automated tests and one-click rollback. Cut incident recovery from 3 hours to 20 minutes (9x faster). Presented forecast outputs to operations managers and walked through edge cases before planning.

Cleaned 100K+ rows of operational data in SQL and Excel, and wrote Python checks that caught malformed inputs before analysis.

Graduated August 31, 2026. Grade 1.81 (German scale, 1.0 best). Thesis on reservoir computing benchmarks (grade 1.3).