Machine Learning Researcher

ML Researcher — Systematic Global Equity

Our client is a systematic global equity investment firm. They build everything themselves — every model, every data pipeline, every piece of infrastructure is written in-house, on Linux, by the people who use it. There's no IT scaffolding and no inherited tooling to lean on. If you join, the systems you build are the systems you'll run.

The Role

Our client is building out a new ML research function alongside their existing systematic equity effort. The mandate is specific: find alpha that traditional factor models — momentum, value, quality, stat arb — structurally can't reach. That means alternative data, nonlinear modeling, and genuine research creativity, not applying off-the-shelf frameworks to standard inputs.

You'll own the full pipeline: sourcing and cleaning raw (often messy) data, building and validating models, and shipping signals into production. This is a research role, but it is not a research-only role — there's no separate team that productionizes your work.

What you'll do

• Research and build nonlinear / ML-based signals for global equity markets, from raw data through production
• Work with alternative and vendor data (equity fundamentals, market data, and less conventional sources) — including the unglamorous work of finding and fixing the errors in it
• Design validation methodology appropriate for financial time series (leakage-aware cross-validation, regime-aware backtesting, etc.)
• Build and maintain your own infrastructure — no dedicated platform team to hand off to

What our client is looking for

• A hard-science foundation: physics, math, CS, statistics, or engineering — undergraduate or graduate.
• Raw programming ability, old school: the ability to write code on a Linux terminal without the aid of agentic AI, Stack Overflow, or code-assistance frameworks — purely from working memory.
• Genuine depth in at least one ML method family, not just familiarity with the term "machine learning." Examples of the kind of specificity meant here: time-series/sequence models (LSTM, GRU), tree ensembles (XGBoost, LightGBM), probabilistic models (HMMs, Gaussian Processes, normalizing flows), or graph-based methods (GNNs) — applied to a real problem, with an understanding of why that method was the right (or wrong) choice.
• Evidence you can defend your work: you should be able to explain not just what a method does, but why you chose it over the alternatives and where it breaks.
• A track record of building things independently — research projects, open-source contributions, or self-directed work that went beyond what was assigned.
• A keen eye for finding the needle in the haystack: you're comfortable with rows and rows of data, and you find real joy in identifying and understanding the cause of bad data. You're relentless about tracking down why data is unreliable — and fixing it.
• A taste for taming messy vendor data: anyone who's worked with vendor data long enough knows they share something in common — most of it is garbage. Are you fired up for taming data from Bloomberg, LSEG, S&P, Barra, FactSet, and countless other vendors? The barrier to entry here should excite you, not discourage you — it's exactly why signals survive where others fail to find them.
• Genuine, unsatisfied curiosity: do you wonder how programming languages are designed in the first place? Using a language is one thing — wanting to be able to design one is another. If you're not satisfied with the status quo and are always looking to challenge the current limits of the technology you use, our client wants to help develop that instinct further.

Nice to have

• A peer-reviewed publication (ML or otherwise) — rigorous, defensible research matters more than the specific field it's in
• Example(s) of personal projects reflecting sincere interest in financial data
• Experience with financial data vendors (Bloomberg, LSEG, S&P, Barra, FactSet) and their quirk

Our client is a small, technically dense team. If the idea of owning a problem end-to-end — with no roadmap and no scaffolding — sounds exciting rather than daunting, our client would like to hear from you.