Writing on ML engineering, system design, and the decisions that don't make it into the code.
8 min read
How moving experiment definitions into YAML config files, rather than code, keeps the training loop stable and makes hypothesis iteration a single command.
April 15, 2026
12 min read
The setup decisions that matter most when training ML models on market data, from data acquisition and target definition through chronological splits, feature engineering, and validation that reflects real trading.
April 22, 2026
10 min read
A walkthrough of the four subsystems in a systematic trading platform: data ingestion, feature calculation, experiment-driven training, and broker execution. Plus the design constraints that shaped each one before the first line of code.
April 29, 2026
11 min read
How a trading platform ingests market data from four sources with different schemas, handles source failures mid-project without touching the rest of the pipeline, detects and fills gaps continuously, and validates data quality before it reaches model training.
May 6, 2026
How the training pipeline enforces feature consistency across training, backtesting, and live inference, why a three-tier model gate is the right checkpoint for this domain, and what fail-fast means in practice when iterating across dozens of experiment configurations.
May 13, 2026
9 min read
Why a single train/test holdout is not enough for financial models, and how three complementary backtest approaches give a more complete picture of model robustness before any strategy reaches paper trading.
May 20, 2026
How the platform integrates with Alpaca for equity trading and Binance for crypto, the validation layer that sits between model signals and submitted orders, and how a local ledger catches discrepancies between what the system intended and what the broker actually executed.
May 27, 2026