Location: In-person, San Francisco (Relocation Supported) Type: Full-time | Competitive Pay + Equity Start Date: ASAP
You will:
- Design agent architectures that reason over technical data and take structured actions
- Build the harnesses that run those agents
- Own evaluation methodology
- Research prompting, fine-tuning, and post-training approaches for reasoning over dense structured technical data
- Study the failure modes of agentic systems in high-stakes deployment settings and design mitigations for them
- Develop calibration and confidence estimation methods so agents know when to act and when to escalate to a human operator
- Publish internal results that other engineers can read, reproduce, and build on
- Shape the research direction of the ML function and, over time, grow and lead a small team
- 4+ years of experience in applied ML, ML research, or a research-heavy engineering role
- Deep fluency with modern language models, agent design, and post-training methods
- Experience building evaluation infrastructure for open-ended tasks where standard benchmarks do not apply
- The judgment to know when an evaluation is telling you something real about production behavior versus when a number moved for reasons that will not hold up
- Ability to work independently on open-ended research problems and communicate findings clearly to engineers who are not ML specialists
- Comfort operating in a domain where the data, the tools, and the evaluation methodology all need to be built
- Prior work on agent harnesses, tool use, or long-horizon reasoning
- Publications or open-source work in LLM evaluation, agent research, or applied ML
Email the following to team@cosmiclabs.io:
Subject line: ML Researcher, Applied ML / [Your Name]
In the body:
- Your name
- Why this role and why Cosmic Labs
- What you bring technically
- Soonest available start date
- Education and years of experience
- U.S. work eligibility status
- Availability (dates and times) for a 15-minute phone call over the next seven days
Attachment:
- PDF resume
Applications reviewed on a rolling basis.