← All open roles▸ RemorooOPEN ROLE · 001
Founding
Engineer.
We're building the autonomous experimentation engine for ML and deep tech teams. You'll help design the systems that let researchers sleep while experiments run.
Full-timeRemote-firstPre-seedEquity-heavyML / Systems
remoroo-session · autoresearch/mar31 ❯ remoroo run --local \
--goal "Follow program.md: improve val_bpb" \
--budget 4
▸ Authenticated as researcher@lab.ai
▸ Run rmo-a8f3 created · Haiku · 4.0h wall time
▸ Baseline: val_bpb = 2.2396 (commit 9138841)
▸ 30 experiments · 8 kept · 22 discarded
▸ val_bpb: 2.2396 → 1.5484 (31% lower)
▸ Verdict: VERIFIED · REPRODUCIBLE
▸ Remoroo is not a coding agent. It's an autonomous experimentation engine for ML and deep tech workflows. You write a spec. You point it at your repo. It runs 30 experiments overnight, evaluates each one against a fixed metric harness, keeps what worked, and discards what didn't. You wake up to a verified patch and a proof trail. The agent didn't guess — it proved.
⟳
Memory system
Demand-paging architecture — pin / page / ephemeral hints, decay, context compression — keeps the agent coherent across 4-hour, 200+ tool-call runs without hitting the context wall.
⌥
AgentLoop v2
A strict plan → edit → train → evaluate protocol. You'll improve reliability, failure recovery, and generalisation to new research workflows beyond autoresearch.
⚑
Process supervision
Detects OOM, NaN loss, silent hangs, and mid-training metric targets from stdout/stderr. False-positive rate has to be low enough to trust. That's real work.
◈
Sandboxed execution
Code runs locally in Docker/venv. The Worker polls the Control Plane, executes jobs, streams results via SSE. Fast, secure, robust — ongoing.
⊕
Cross-run memory
recall and record_lesson give the agent institutional knowledge across runs. You'll improve retrieval quality and how the agent builds on its own history.
We’re looking for a strong technical builder who also has the social instincts of a founder: someone who can listen deeply to customer pain, explain complex ideas clearly, handle skepticism with confidence, and build trust quickly with users, teammates, and partners. This role requires clear communication, sharp thinking, and the ability to turn technical credibility into momentum for the company.
Technical mustsAbsolute must
LLM context, tool calling & agent memory
This is the core of everything we build. You understand how context windows fill up, how tool call loops fail, and how memory systems keep agents coherent over long runs. You've thought hard about these problems — not just used an SDK.
Absolute must
You build with coding agents fluently
Claude Code, Cursor, Devin — these are your daily tools, not novelties. You know how to prompt effectively, where agents break down, and how to structure work so an agent can execute it reliably. We build for people like you.
Absolute must
You've shipped something with agents
Memory, tool use, long-horizon planning, evaluation — something non-trivial. You know what makes agents fail over time, and you have something to show for it.
Absolute must
Machine Learning/ Robotics/ Autonomous Systems
You have a strong background in machine learning, robotics, or autonomous systems. You understand the challenges and opportunities in these domains and how to apply them to the problems we're solving.
Nice to have
- Docker, Firecracker, or sandboxed execution environments
- A knack for user interface design
No whiteboard. No algorithmic trivia. We want to see how you think about systems that fail in slow, subtle ways.
STEP 01
First interview
Short intro conversation. We read everything you send carefully.
STEP 02
Practical project
Real problem, 4-8 hours.
STEP 03
Systems conversation
Agent design, memory architecture, failure modes.
STEP 04
How you work
Founding team fit, how you operate without structure.
Apply now
Include a GitHub repo for a project you've worked on.
or a short Loom/YouTube video (2–5 min)