The company is an AI quantitative finance lab that works on helping machines understand markets better.
Current models can explain finance and write research code, but they struggle to form, test and revise reliable market views. The team is building agentic quantitative-research systems, expert-graded evaluations and reinforcement-learning environments to close that gap.
The founder, Jeremie Cohen, previously worked as a quantitative portfolio manager at WorldQuant and as lead data scientist at Millennium. He started the company after seeing that models can discuss markets but cannot yet run rigorous, point-in-time research that holds up through execution, backtesting and out-of-sample testing.
The team is small and moves quickly. You would own the post-training research agenda from an early stage, run fast experiments with a lot of technical independence, and train on executable research trajectories, expert demonstrations and feedback from quant experts rather than static financial text alone.
Responsibilities
Own the research and engineering needed to make models better at complex, multi-step quantitative-research work.
Cover training-data design, supervised fine-tuning, preference optimization, reinforcement learning, verifier building, evaluation and close error analysis.
Post-train open-weight models using SFT, DPO or similar preference methods, reinforcement learning and verifier-based rewards.
Build training sets from research trajectories, expert demonstrations, executable tasks, critiques, rejected paths and structured feedback.
Train agents to form hypotheses, express them mathematically, write and run code, assess results and rework failed approaches.
Design rewards and verifiers that check point-in-time correctness, statistical validity, reproducibility, code execution and research quality.
Build evaluations that separate real capability gains from overfitting to benchmarks.
Run ablations and detailed error analyses to find which data, objectives and training methods drive progress.
Improve distributed training, inference, data-generation and experiment-management infrastructure.
Partner with the quantitative researchers to turn institutional research workflows into useful training environments.
In the first 30 days, learn the research harness, set baseline performance for selected open-weight models, and build and benchmark a verifier for a mean-reversion task covering temporal leakage, execution correctness, statistical validity and reproducibility.
By 60 days, complete a first post-training run, compare it against the baseline, and pinpoint the highest-leverage failures in data, reasoning, tool use and reward design.
By 90 days, run reinforcement-learning experiments aimed at agentic hypothesis generation and research execution, and report what improved, what failed and the next research plan.
Requirements
Hands-on experience post-training open-weight language models, not only prompting models or building applications on closed APIs.
Solid knowledge of SFT, preference optimization, reinforcement learning, reward modeling, distillation and synthetic-data generation.
Experience designing datasets and feedback signals that produce measurable changes in model behavior.
Proficiency in Python, PyTorch and current open-source model tooling.
Experience running and debugging substantial training experiments.
Sound experimental judgment across evaluation design, ablations and failure analysis.
Ability to move from research ideas to reliable implementation.
Comfort owning ambiguous problems on a small, fast-moving team.
No finance background is required. The team already has deep quantitative-research expertise and needs someone who understands how training changes model behavior.
Any experience level; new graduates are welcome.
Especially relevant experience:
RL with verifiable rewards or process-level feedback.
Post-training reasoning, coding, mathematical or tool-using models.
Agentic systems operating over long, multi-step trajectories.
Distributed model training and inference.
Data curation, synthetic-data pipelines or rejection sampling at scale.
Evaluation harnesses, execution sandboxes or model-based graders.
Time-series modeling, backtesting or point-in-time datasets.
Published research, meaningful open-source contributions or strong independent experiments.
Prior finance experience is welcome but not required.
Conditions
Candidates must be US citizens or hold a visa.
Where you’d work
Fully remote
You can work from
United States
United Kingdom
Canada
Europe
France
Bulgaria
Hungary
No visa sponsorship
You must already be able to work in the United States
Good
Your chances
Still hiring, not crowded yet, and a person reads your message.
19 checks run
6 good signs
2 red flags
Still hiring?
12 checks
Actively hiring
StaleHiring
In its favour3
Still on the company's own careers site, checked 1 h agoModerate evidence
Posted 2 days ago: newer than 94% of open rolesModerate evidence
A hiring contact is attached to itSlight evidence
Against it1
Very wide salary range: $150K to $350KSlight evidence
How crowded?
7 checks
Low
QuietCrowded
In its favour3
Open for 2 days: you'd be among the earlier applicantsModerate evidence
You can message the hiring contact and skip the queueModerate evidence
Only for people already authorized to work in United StatesSlight evidence
Against it1
Pays more than 84% of similar roles: that draws applicantsSlight evidence
?
Fits Me
How well does this role fit you?
Answer a few questions or drop your CV, and every role gets a fit score with the reasons, this one first.