HFT Architecture builds short-horizon price prediction and the trading system around it. Models are built by a pipeline that turns order-book records into fitted, validated models, one market and one horizon at a time. This role works on that pipeline next to more experienced engineers and learns the trade on real market data. It is fully remote: you can work from anywhere in the world.
What you will do
- Prepare order-book data for training: cleaning, alignment in time, features, and checks that the data is what it claims to be.
- Run experiments end to end and write down what you ran, on what data, and what came out.
- Find leakage. Learn where the future gets into a training set, and add the check that keeps it out.
- Reproduce a result somebody else got, and say where yours differs and why.
- Take reviews of your code and your numbers, and review others' in turn.
What you need
- Python you write every day, with tests.
- The basics of machine learning that you can explain, not only call: a train and test split, overfitting, cross-validation, and why ordinary cross-validation breaks on time series.
- Enough statistics to doubt a good result before you celebrate it.
- One project of your own, a course project or a competition entry, that you can show and talk through: the data, the model, what went wrong.
- Writing clear enough that someone else can repeat what you did.
Good to have
- Any time series work: prices, sensors, logs.
- pandas or Polars on data larger than memory.
- A Kaggle notebook, a GitHub repository or a write-up of yours that somebody else used.
Read before you apply
This article shows how we measure and how we write it up: Why Your Market Maker Loses Money. The application form asks you to say, in one sentence, what one of our articles is about.
How we hire
- You apply on this page, with a link to your work.
- A conversation about a project of yours: what you did, what you would change.
- One practical task on real data, small enough to finish in an evening.
- A decision, with the reason.