At Tata Consultancy Services, I engineered enterprise-grade LLM guardrails that reduced unsafe model outputs by 45%, with risk detection accuracy above 92%. I used embedding-based and transformer-based classification to identify risks.
I developed LLM and Generative AI applications using RAG, multi-agent systems, and evaluation loops with LangChain and LangGraph. I also worked with vector databases to support scalable, explainable deployments.
I implemented an AI-agent solution using Azure OpenAI models to automate an NFR evaluation workflow. Structured chunking and workflow orchestration improved processing efficiency by 50% and reduced manual effort.
In my projects, I built Flask-based microservices for LLM guardrails and for evaluating trust factors such as resiliency, explainability, bias, and model sufficiency. I also addressed security vulnerabilities flagged by SAST, CVC, and SonarQube to support compliance across AI pipelines.

