recsys-experiments

Run and extend the repository's recommendation-system projects covering the ranking funnel, two-tower ANN, point-in-time feature consistency, and online serving reliability.

SKILL.md

GitHub

Recommendation experiments

  1. Choose one project under templates/ based on the learning objective.
  2. Read its README and cited PDF pages before changing code.
  3. Run the baseline, then the documented fault injection.
  4. Change one variable at a time; record input, version, metrics, and expected invariant.
  5. Validate the consumer-visible outcome, not internal wiring.
  6. Treat every [MOCK] as a teaching substitute. A production proposal must add real integrations, capacity, timeout, degradation, observability, versioning, rollback, and offline/online parity.