About

I'm Siddharth Sukumar. I build agent systems for production — specifically, ones that have to work reliably without a human in the loop on every step.

Remediate Labs is the latest version of that work: a multi-agent platform that watches a production deployment, catches errors, diagnoses them against the actual codebase, and opens fix PRs. It's been the centerpiece of my engineering work for the last several months. The blog covers what I've learned shipping it.

What I'm working on

  • agent-platform — a 7-agent incident remediation pipeline on FastAPI, Postgres and ECS Fargate. Diagnosis runs on the open-weight DeepSeek-V4.1-Flash with a harness evolved by an optimizer agent (78.8% → 89.9% localization on repos it never saw), and every change to it passes a statistical regression gate. Architecture deep-dive →
  • Writing — the 7-part Self-Improving Harness series, including what didn't work, plus retrieval, code-graph and cost posts on the blog.

What I look for

I'm interested in AI Engineer / agent systems roles where the work involves: production-grade agent design, harness engineering and self-improving agents, eval-driven decisions on model and prompt changes, retrieval architecture, and observability for non-deterministic systems.

Get in touch