Research Fellow
About the company
Our client is an early-stage AI research lab building benchmarks, reinforcement-learning environments and training data that they sell directly to frontier AI labs rather than to a single exclusive partner. Founded in 2024, they are around ten people today and expect to roughly triple within the year, so this is joining in the single digits. The founding team comes from self-driving, large-scale search and frontier-model research, and the company is backed by angel investors who work at the leading AI labs.
The role
This is a fixed-term research fellowship — roughly six months, flexible around academic semesters — aimed squarely at mid-PhD and recently graduated researchers who want to test the industry waters without leaving their program behind. The day-to-day work is the same as that of the lab's full-time researchers: designing tasks and environments that are easy for a human to verify but hard for current models to solve, largely long-horizon agent work and increasingly AI security, then shipping that work to a frontier lab within weeks. Two things to be clear-eyed about before applying: this is a 1099 contract rather than employment, and it carries no equity. Strong fellows are an explicit target pool for a full-time offer afterwards.
What you'll do
- Read the current literature, pick a research question and drive it to completion yourself
- Design long-horizon tasks and reinforcement-learning environments that a human can verify but current models find hard
- Build the software for your own research — this is a hands-on coding role, not a supervisory one
- Work increasingly in AI security: target development, vulnerability discovery and sub-agent exploitation loops
- Ship finished research and environments directly to frontier AI lab customers, on a timescale of weeks
What we're looking for
These are hard requirements. Check each one that applies to you.
Don't check every box? No worries — submit your info here and we'll keep you in mind for roles that could be a better fit.
Submit Your InfoNice to have
- Experience with advanced post-training, distillation or high-stakes evaluation methodology
- A background at a university known for AI research, or an advisor known for this kind of work
- Depth on how specific frontier models actually behave — their quirks and failure modes, not just using an LLM as a tool
- Real open-source benchmark contributions from outside the usual tier-one pipeline — the client values these highly and is explicit that it does not screen on school name
Tech stack
Python, with reinforcement-learning environments, agent frameworks and evaluation harnesses. The research areas are long-horizon agent tasks, benchmarking, AI safety and alignment, and AI cybersecurity.
Compensation & logistics
- $150k–$200k annualized equivalent, paid monthly — roughly $12,500 to $16,667 a month
- Engaged as a 1099 contractor rather than a W2 employee, with no equity
- Term is roughly six months, flexible around academic semesters
- On-site 5 days a week in San Francisco or Dublin, Ireland — this is not a remote or hybrid role
- Open to visa transfers such as OPT and H-1B — worth confirming your own situation early, since the 1099 structure can affect what is possible
- 0 to 1 years of full-time AI research experience — this is a fellowship, not a senior research seat
- Up to 4 fellows planned
- Four interview stages: a recruiter screen, a self-paced 48-hour take-home, a 30-minute deep dive with a founder, and a half-day on-site