Devanshu Chicholikar
Software Engineer · AI Engineer · MCP, RAG, evals
Boston, MA · Open to Software Engineer, AI Engineer and Forward Deployed Engineer roles
Hey! I ship AI systems to production end to end: MCP servers, RAG with retrieval evals, LLM-as-a-judge eval harnesses and voice agents, plus the full-stack, infra and design work around them. OpenCodeIntel is a code-search platform for AI coding agents (web app, API and MCP server), and Overhear grades every call a voice agent takes: code checks the call log, and an LLM judge checks what the agent said against the clinic's real open slots. MS Software Engineering Systems from Northeastern, finished May 2026.
Before that I spent two years full-time building a Java / Spring Boot quotation platform, and TA'd Network Structures & Cloud Computing for 100+ grad students across two semesters. Teaching set my bar: if I can't explain it at 11 PM to someone whose AWS setup is on fire, I don't understand it yet.
I build systems that work reliably under pressure. Distributed systems. Cloud infrastructure. APIs that respond fast. UIs that people can actually use.
Half the job never makes the resume: figuring out what the customer actually needs, saying no to the clever version, and shipping the one that works.
If it isn't measured, I don't trust it, my own work included. That's why every project here comes with its numbers and its caveats.
- You're paying for tokens the model isn't reading.
- The model isn't smarter. It's just enough. And "just enough" is what actually ships.
- Scale was never the moat.
- Some rerankers make code search worse. I know because I tried one, measured it, and wrote it down.
- If code can check it, don't ask an LLM. In Overhear, code decides whether the caller was verified before anything got booked. The judge only rules on what the agent said.
- "I don't know" is a feature. A confident wrong "in stock" sends someone to an empty pharmacy shelf.
- My eval set is 39 calls, mostly synthetic, one labeler. That sentence sits right next to the F1, where it belongs.
- Pineapple belongs on pizza. This one isn't data-driven.
- Discipline
- Software + AI engineering · forward deployed
- Layer
- MCP · RAG · evals · voice agents
- Stack
- TypeScript · Python · Go · AWS
- Retrieval
- BM25 + vectors, RRF fusion, rerank
- Shipping
- OpenCodeIntel · Overhear · CallBudget · Saar
- Based in
- Boston, MA
- OpenCodeIntel: Code search for AI coding agents, so they stop guessing
- Overhear: A QA analyst for voice agents. It listens to every call so you don't have to
- CallBudget: Teach a pharmacy-finder to call less and find more
- Saar: A Claude.ai token and cost meter that runs entirely in the browser
- Rerankers made my code search worse. Then the same one made it better.
- My LLM judge missed a skipped ID check. 24 lines of code didn't.
Graduate Teaching Assistant, CSYE 6225 Network Structures and Cloud Computing
Northeastern University · Sep 2025 - May 2026 · Boston, MA
Software Engineer, B2B Industrial-Equipment Platform
Jaksh Enterprise · Aug 2022 - Jul 2024 · Ahmedabad, India
Software Development Engineer Intern
Pitney Bowes · Jan 2022 - Jul 2022 · Pune, India
M.S. in Software Engineering Systems
Northeastern University · Sep 2024 - May 2026
B.Tech in Information and Communication Technology
Dhirubhai Ambani Institute of ICT · Aug 2018 - May 2022
Email chicholikar.d@northeastern.edu, or find me on LinkedIn and GitHub.