The Anatomy of an Machine Learning Powered Stock Picking Engine

An ML-powered engine is a complex beast, more so if it's targeting one of the toughest tasks some humans undertake: squeezing out money from reluctant markets. I'll talk about the philosophy behind the engine, and why the task is simultaneously more complicated and less esoteric than it appears.

In particular, I'll talk through the following topics:

‍
* Markets - Patterns, Regimes, Regime Shifts, and Sentiment
* Engine Architecture - Tech stack, environment, execution, scheduling
* Data ingestion - complexity, taxonomy, quality, transforms, backups
* Some thoughts on feature engineering
* ML Modeling - forecasting, regime detection, explanations
* ML Ops - Workflow monitoring, feature explorer, model diagnostics, performance analytics
* Tackling performance bottlenecks
* How the engine performed in out-of-sample testing

You will find that, at the core, the tasks above are what you will have to undertake in order to build in this space, beyond the particulars of the asset class you're targeting (equities, commodities, currencies, or crypto).

‍