AI Engineering at Trading Firms
Production ML versus research, what skills get hired versus what gets ignored, and the comp ranges nobody publishes.
By Colin van Eck
Almost every trading firm we work with is now hiring for "AI" or "ML" roles. The titles cluster — ML Engineer, ML Researcher, Applied Scientist, Quant ML Lead — but the actual work behind the titles varies wildly, and so do the candidates who succeed in them.
Here's how we've come to think about the split.
Production ML vs. Research ML
The two roles look identical on a job spec and almost nothing alike in practice.
Production ML engineers spend most of their week on infrastructure: feature stores, model serving latency, monitoring, retraining pipelines, the boring-but-load-bearing work of getting a model from notebook to live trading. The candidates who thrive here have deep software engineering chops — distributed systems, low-latency C++/Rust, profiling tools — with enough ML literacy to understand what the model is actually doing in production.
Research ML scientists spend most of their week doing what looks more like academic work: reading papers, running ablations, structuring experiments, writing internal memos. They tend to have stronger statistics / theoretical ML backgrounds, often with a PhD in ML, applied math or a hard science.
The mistake we see firms make repeatedly: hiring one when they needed the other. A production engineer dropped into a pure research role gets bored and leaves in 6 months. A researcher dropped into production work fails to ship anything that survives contact with live latency budgets.
The fix is brutal honesty in the spec about which one you actually need. We'll push back on this hard during intake — it saves everyone time later.
What gets hired vs. what gets ignored
Across roughly 30 ML/AI searches we've supported in the last 18 months, the skills that consistently move candidates to the top of shortlists:
- Real production deployment experience. Anyone who has actually shipped an ML model into a live PnL stream — even a small one — has a meaningful edge over candidates whose work has stayed in notebooks.
- Latency awareness. Trading firms care about microseconds; most ML candidates are accustomed to pipelines measured in seconds or minutes. The ones who can talk credibly about quantising models, optimising inference, or designing the boundary between ML and non-ML execution stand out immediately.
- Statistical rigour. Particularly in roles touching alpha research. The strongest candidates are sceptical of their own results, run honest ablations, and know what backtesting overfitting looks like in the wild.
Skills that rarely move the needle, despite candidates over-indexing on them in CVs:
- Generic LLM / GenAI experience. Trading firms with serious LLM use cases are rare. Most claims to "extensive GenAI work" amount to API calls.
- Kaggle competition rankings. Useful as a directional signal, not as a hiring criterion.
Comp ranges (rough, Amsterdam / London)
For senior production ML engineers at established prop firms, total comp now sits roughly:
- Mid-level (3–5 years): €140–200k all-in
- Senior (6–9 years): €220–340k all-in
- Lead / Staff: €350–500k+ all-in
Research ML scientists trend slightly higher at the top end, particularly when there's a clear PnL attribution path. Crypto-native firms add another 20–30% across the board for the right hire.
These numbers move quickly. If you're hiring or thinking about a move, get in touch and we'll give you the current read on your specific level and seniority.