Join a small, fast-moving team where you’ll explore, invent, and discover.
We’re training foundational models for financial applications on unique, multimodal datasets.
What you’ll do
- Formulate and pursue original research agendas (reasoning, tool connectivity, robustness, alignment, data/compute efficiency).
-Follow Ideas to PoC to MVP to training to validation, owning all steps of the process with deep collaboration with Research Engineers (see other open role).
-Build rigorous evaluation suites through simulated environments and real time evaluation.
-Prototype fast with real data; partner with Research Engineers to take promising ideas to production.
You might be a fit if you
- Are a self-starter who thrives on coding autonomy and project collaboration within small-team, high velocity environments.
-Have strong ML research chops (theory and practice), and write clean, reproducible code (e.g. PyTorch/JAX).
-Think in systems: data quality, scaling laws, evaluation, and deployment constraints - not just model internals.
-Prefer clear principles, low ego and prefer collaboration over titles and politics. Helpful backgrounds (any of the following) - Frontier labs experience (e.g., Anthropic, OpenAI, DeepMind); or
- Novel discovery through published work (NeurIPS/ICML/ICLR); or
- Led research in an applied setting (startups, open-source, or products at scale); or
- Standout independent achievement (notable OSS, benchmarks, or widely used methods).
Research areas we’re excited about
- Efficient training (HRM, distillation, sparse/MoE, FSDP, etc.).
-Reasoning and tool use (automated trading and agentic interaction with data sources).
-Multimodal Inputs and Outputs (multiple forms of text and time-series both in and out).
What we offer
- A well-funded trading firm expanding into AI research - your ideas set direction and standards.
-Real ownership from hypothesis to deployment.
-Competitive base with meaningful upside tied to research impact.
-A culture optimized for deep work, fast learning, and doing the right thing.