ML Engineering for Quants

Read these first

1. 01Architecture of a Quant ML Stack

   A working quant ML stack is less about which model library is used and more about a handful of infrastructure guarantees, point-in-time feature computation, reproducible training snapshots, and consistent inference, without which even a correct model produces wrong answers.

2. 02Serving Models Inside a Trading System

   A model that scores brilliantly in a research notebook still has to survive being called thousands of times a day inside a live trading system, under latency budgets, version-control discipline and failure modes that research code was never designed to handle.

3. 03Testing Machine-Learning Code

   Unit tests catch broken code; they do not catch a model that runs fine and learns the wrong thing. ML testing needs a second layer aimed at data and behaviour, not just syntax.

Then the rest

1. 04C++ Inference for Tree Ensembles
2. 05Dataloader Throughput and GPU Utilisation
3. 06Designing Feature Pipeline DAGs
4. 07Distributed Training on Panel Data
5. 08Hot-Reloading Models Without Downtime
6. 09Immutable Training-Set Snapshots
7. 10Incremental vs Full Feature Recomputation
8. 11Inference Latency Budgets
9. 12Model Rollback and Freeze Procedures
10. 13Nondeterminism in GPU Training
11. 14Offline/Online Feature Parity
12. 15ONNX and Cross-Language Model Export
13. 16Packaging and Serialising Model Artefacts
14. 17Point-in-Time Feature Computation Engines
15. 18Quantisation and Pruning for Latency
16. 19Running Hyperparameter Sweeps at Scale
17. 20Sweep Result Databases and Model Leaderboards
18. 21Unit Tests for Feature Transforms