[Remote] Machine Learning Engineer

Note: The job is a remote job and is open to candidates in USA. Deeter Analytics is a privately held investment research and trading firm managing its own capital across public markets and combining human judgment with modern AI. The Machine Learning Engineer will build deep-learning models for market research, manage small-scale GPU infrastructure, conduct rigorous experiments, and take ideas from raw data through production systems that inform trading.

Responsibilities

• The project. A new research effort with a senior lead; from a desk idea and raw data to models, experiments, and a result the team can act on. You own the loop end to end
• The models. Deep-learning models on market data built, trained, ablated, and improved by you, from the first baseline to something that runs every day
• The compute. Your own small GPU setup, local or on AWS: environment, drivers, containers, storage, experiment tracking, cost. Small-scale by design; you keep it running and you keep it cheap
• The evaluation. Leakage-proof validation on time-ordered data, regime-aware testing, honest baselines; knowing the difference between a result that's real and one that's noise, and being able to show which is which
• The reading. Recent work in foundation models, time-series, and RL: read it, reproduce what matters, and write up what you found in a page the team will act on
• AI leverage. Use modern AI tools to move faster (code, literature, data wrangling) and check their work

Skills

• The role is full-time and fully remote, US-based
• Grounded in fundamentals. You know what's inside the models you train (optimization, initialization, normalization, attention, why a run diverges or plateaus) and the math underneath: linear algebra, probability, statistics. You can derive the gradient of a loss and say what changes when the batch size doubles
• A builder. Personal projects, hackathon builds, a model trained on your own machine, a repo people actually use. You've shipped things nobody assigned you
• Scrappy and hands-on. You'd rather stand up the box, fix the CUDA driver, and get the first experiment running tonight than wait for someone to provision it
• Honest about results. You go looking for the reason your number is too good before anyone else does, and you'd rather kill your own result than have the market do it
• Low ego and coachable. You take feedback well, update quickly when the facts change, and care more about the answer than the credit
• Curious about markets, not credentialed in them. Interest helps; experience isn't required; the lead has that, and we'll teach you the domain
• End to end. Data, model, infra, evaluation, write-up: you own the loop, not a slice of it
• Clean experiments. Versioned data, seeds, ablations, honest baselines; every claim comes with the run that backs it
• Fast and exact. You get to a first result quickly and don't let rigor slip when you do
• AI-native. Fluent with modern AI tools for code, literature, and data work; you get real leverage from them and you verify what they give you
• Self-directed. You thrive working remotely with low guardrails, managing your own time and flagging what needs attention without being asked
• Clear in writing. A page that says what you tried, what happened, what it means, and what's next
• Deep learning fundamentals. Optimization, regularization, sequence models and attention, evaluation, and the linear algebra, probability, and statistics underneath
• Python and PyTorch. Strong and idiomatic, from first principles when a library doesn't fit; NumPy and pandas for the data work around it
• Small-scale GPU infrastructure. Setting up and running your own training and inference environment on a local machine or a few AWS instances: CUDA, containers, storage, monitoring, cost control
• Time-ordered data. Working with data that has a clock: splits that don't leak, backtest hygiene, distribution shift
• Reproducing research. Reading a paper, getting it running, and knowing where it breaks on your data
• Bonus, not required. Fine-tuning or serving LLMs on your own hardware; CUDA or Triton; time-series forecasting

Benefits

• A seat inside a live trading operation, working directly with the traders and researchers who act on your models.
• A deliberate growth path: own one project end to end first, then take on more of the research agenda as you prove out.
• A small, low-ego, fully remote team.
• Bonus

Company Overview