Sarath Chandra Bellam

I build agentic AI systems that have to survive production 🤖

I'm a data scientist working on agentic AI for enterprise. I started out as a mechanical engineer and found my way to language models — what carried over wasn't the math but systems thinking: tolerances, failure modes, what happens at the edges.

These days I spend most of my time on the unglamorous parts of AI that decide whether it survives production: context efficiency, evaluation, tool and MCP design, and knowing when an agent should simply stop and ask. I care more about using the tokens you have well than about having more of them.

Timeline

2026 —

Associate Director, Data Science at S&P Global. I design and ship agentic systems for one of the largest financial-data companies in the world — where being wrong is expensive and "it usually works" isn't an answer. Building an internal agent framework, an MCP playground, and an AI workspace for analysts who don't write code.

2024 — 26

Lead Data Scientist at S&P Global. Led R&D on generative and agentic AI, using the Model Context Protocol and LLMs to bring enterprise applications from prototype to production.

2022 — 24

Senior Software Engineer at S&P Global. Built conversational AI with Rasa and deep-learning systems that had to run for real users.

2021 — 22

Software Engineer at S&P Global. Rasa, NLTK, and the NLP pipelines underneath them.

2019 — 21

Associate Data Scientist at Cognizant. Data processing, NLP, and my first ML systems in anger.

2016 — 19

Software Engineer at Tech Mahindra. Python developer — the foundation in building things that don't fall over.

2012 — 16

B.Tech, Mechanical Engineering at Gudlavalleru Engineering College. Sports captain. Where systems thinking started — long before it had anything to do with AI.

Writing

The MCP Revolution: How Tool Granularity Can Make or Break Your AI's Performance and Cost — Medium, 2025 Context Relevance to Context Efficiency: The Rise of Context-Window Architecture — Medium, 2025