NEW: The Blockchain Data Engineering Guide is out →NEW: The Blockchain Data Engineering Guide is out →

Research Scientist

Tokka Labs logo

Tokka Labs

📍 Singapore, Singapore💰Competitive🕐 Posted
Data ScientistOnsitedefi
pythonpytorchllmreinforcement-learningmachine-learningtransformers
ApplyTailor my resume for this job

Free fit check: see how your resume matches this job

Job Description

About Tokka Labs

Tokka Labs is a Singapore-based proprietary trading firm and market maker in digital assets and tokenised real-world assets, bringing traditional instruments into markets that never close.

Our team has built and run trading businesses at top-tier global firms, and we bring that rigour to on-chain markets, where continuous trading and evolving market structure create problems with no existing playbook.

We build our systems and strategies in-house and stay deliberately lean and flat. Traders, researchers and engineers work side by side, and good ideas reach live trading quickly. With AI and agentic tooling central to how we work, each person has the leverage of a much larger team and focuses on the hardest problems. We look for curiosity and ownership, and give strong people the room to act on their ideas and see their impact directly.

The Role

We're hiring a research scientist to build AI systems that learn from their own research and improve over time, and to apply them to real problems across the firm.

We believe the most durable edge comes from knowledge that compounds. We're building AI-driven research systems that help our teams find and validate improvements faster. Your job is to make them general, reliable and cumulative, and bring them to new problems across the firm.

You'll report directly to the head of research and work with our trading and engineering teams. The core challenges are open research problems: getting automated research to build on what it has learned, and making sure the improvements it finds are real.

We care more about strong machine learning and evaluation judgment than finance knowledge.

What You'll Do

  • Research: Find where current frontier models fall short as autonomous researchers in our systems, and turn those limits into research directions.
  • Experiments: Design and run experiments, and turn the results into new methods.
  • Knowledge: Design how our systems capture proven methods and tools from their work, and reuse them so each new problem is solved faster and at lower cost.
  • Evaluation: Build evaluations that tell real improvement from noise, and what it costs in time and tokens, across ML, software engineering and strategy problems.
  • Generalize: Turn our research systems into reusable tools that new problems can adopt quickly.
  • Automation: Reduce the manual work of bringing a new problem on board.
  • Operations: Keep the systems reliable, observable and cost-effective.
  • Communication: Explain what worked, what didn't and why.

Requirements

  • Research track record: Publications or equivalent research in ML, optimization, statistics or a related field, and the ability to turn papers into ideas worth testing.
  • Agents and learning systems: Hands-on work with frontier LLM agents, reinforcement learning or systems that learn from experience, including a clear sense of where they break.
  • Rigor: You design careful experiments and catch your own overfitting or misleading results.
  • Applied ML: Hands-on experience training and evaluating models on real data, ideally shipped to production.
  • Coding: You can turn ideas into working production code.
  • Background: A PhD with 1+ year of experience, or a Master's with 3+ years, in Computer Science, Statistics, Electrical Engineering, Applied Mathematics, Physics or a related quantitative field, or equivalent research experience.
  • Clear communication: You can explain a result, and its uncertainty, to people who weren't in the weeds.
  • Ownership and resilience: You'd rather own a problem end to end, across team boundaries, than hand it off. You untangle ambiguity until you understand the real problem, and you stay resilient when experiments fail or plans change.
  • Pace and adaptability: You thrive in fast-paced environments that require you to think on your feet and drive solutions within short time frames. You juggle multiple priorities, pivot quickly when they change, and do your best work when they shift.

Nice to Have

  • Self-improving systems: Experience with automated scientific discovery, open-ended search or self-improving agents.
  • Time-series data: Experience modeling noisy, nonstationary time series.
  • Modeling range: Hands-on PyTorch, including transformers or other sequence models.
  • Markets: Exposure to market microstructure, crypto, DeFi or arbitrage.

Why Join Us

  • Build with real backing. Leadership is investing heavily in AI and agentic tooling as core strategy, not a side project.
  • Work alongside people who have built and run trading businesses at top-tier global firms.
  • Real stakes. Your code runs on the critical path, and its impact shows up directly in P&L.

Unchain Data provides Web3 data job aggregation as a common good. Jobs are posted by third parties and are not individually verified. Always exercise caution: never download software requested during a hiring process, avoid clicking unfamiliar links in interviews, make sure to verify URLs are legit, and use trusted meeting tools like Google Meet or Zoom.

Want to land this role?

Practice SQL on real blockchain data. Build the query skills crypto data teams test for in interviews.

Explore WizardCamp