Deepank A.

About Deepank A.

I build AI agents that actually work in production.At Atlas, I work across the full stack of agent engineering — designing autonomous customer support agents that drove response accuracy from 40% to 95%, architecting RAG-based knowledge pipelines at scale, cutting inference costs by 40%, and building internal developer tooling powered by AI. Most of my time goes into making LLMs reliable, fast, and useful for real customers.I tend to gravitate toward ambiguous, zero-to-one problems — Hermes, our internal agentic dev tooling, started as an idea I pitched and built end-to-end. I pick things up quickly and context-switch between prompt engineering, infrastructure, and full-stack work without friction.Before the AI world, I spent two years at Quadeye building high-frequency trading simulations in C++, optimizing runtimes by 30-35%, and stepping up as project lead when the role unexpectedly opened up.I studied CS at IIT Kharagpur (9.47 CGPA), where I apparently peaked at competitive programming (Google Kick Start Rank 39, ACM ICPC Rank 50).When I\'m not prompting LLMs or arguing with TypeScript, you\'ll find me on the badminton court, at the gym, or hunting for good coffee.https://topmate.io/deepank_agrawal

Experience

1. Software Engineer, Agents

   Atlas

   Jun 2025 — Present

   Working across AI agent engineering, knowledge pipelines, dev tooling and full-stack product work.

Education

• Indian Institute of Technology, Kharagpur

  BTech + MTech, Computer Science and Engineering

  2017 — 2022