Harish Balaji

I build production-ready GenAI systems with a focus on LLM/RAG architecture and dependable evaluation. I care about translating model behavior into measurable quality, so teams can trust outputs and ship faster.

At WABTEC, I designed and deployed a multi-document RAG chatbot using LangChain and Chroma, working with 300–500 service manual documents and improving retrieval speed by 35% versus manual search. I also created an LLM-based evaluation framework across 7 quality dimensions—targeting >90% composite quality.

Earlier, at Indium Software, I applied LLMs to quality engineering workflows—improving test coverage by 30% through LLM-powered test case strategies and accelerating bug triage with a semantic similarity and anomaly-scoring pipeline. I strengthened AI-driven QA validation by integrating autonomous agents with LLM backends for test automation and error detection.

I’m also actively publishing and building agentic systems, including a Routledge 2025 chapter on agentic AI integration and projects like a multi-agent water quality prediction pipeline and an agentic precision agriculture decision engine with explainability for non-technical users.