Machine Learning Engineer

• Build the pipeline from raw order-book records to a fitted, validated model: data preparation, features, training, out-of-sample measurement.
• Make every run reproducible: the same data and the same configuration give the same model and the same numbers.
• Keep the future out of the training set: leakage controls and embargoed walk-forward validation built into the pipeline, where nobody has to remember them.
• Bring the pipeline to a new market. A new market still needs its connector and its features adapted.
• Measure what a model costs to run, and get it to where the trading system can use it.

• Machine learning systems you built and ran yourself, from data to a model somebody depended on.
• Python, and the engineering around it: tests, versioned data and configuration, runs you can repeat a month later.
• Time series handled without leakage. You can say where the future can get into a training set and how you close each door.
• Enough statistics to tell a real improvement from a lucky split.
• Large data on ordinary hardware: order-book records come in tens of millions of rows a day.

• Market data, tick or order-book.
• Low-latency inference, or a model moved out of Python into a production path.
• A pipeline of yours that produced a wrong model quietly, and how you caught it.

Two articles show how we measure and how we write it up, and the form asks about one of them: Why Your Market Maker Loses Money and Latency Arbitrage Won 30 Days of 30 and Lost to Fees.

• You apply on this page, with a link and an answer to one question about the article.
• A technical conversation about a system you built.
• One practical task on real data, small enough to finish in an evening.
• A decision, with the reason.