dimamitya

software engineer · ai agent systems

i build reliable systems for AI agents.

harness juggler. rl environments, orchestration, agent-safety, and the eval harnesses that keep fleets of agents honest.

// about

about

i build the systems ai agents run on: rl environments, eval harnesses, multi-agent orchestration, and the reward design and agent-safety that keep them honest. reward that resists goodhart; safety that's architectural, not just prompt-level.

before that, three years shipping web and smart contracts. i like systems that are boring enough to trust: explicit over implicit, signal without noise, the minimum code that solves today's problem.

// skills

skills

languages

  • rust
  • python
  • typescript

ai / agents / eval

  • rl-environment design
  • eval-harness design
  • reward & grader design
  • anti-reward-hacking
  • agent-trajectory data
  • multi-agent orchestration
  • agent-safety architecture
  • llm evaluation

tools

  • claude code
  • codex
  • gemini cli
  • harbor
  • terminal-bench

backend / systems

  • fastapi
  • async / concurrent
  • postgres
  • redis
  • distributed systems
  • docker
  • prometheus

infra

  • fleet automation
  • provisioning clis
  • ci/cd
  • linux

// work

selected work

// activity

activity

14,941 contributions in the last year

last year on github ↗lessmore

// writing

writing

all posts →

// work with me

work with me

pick what fits, then book below.

embed not loading? open the booking page ↗

// contact

contact

the fastest ways to reach me: