From Inference to Features: Build a Core ML Platform from Scratch
summary
Hands-on build of five concrete components — feature store, model registry, orchestrator, inference engine and basic monitoring — using Test-Driven Development, Domain‑Driven Design and hexagonal architecture. Presenter live‑codes in Python with in‑memory repositories, a message bus of commands/events/errors, handlers and translators, shows unit tests passing and an end‑to‑end scenario where missing models/features are recovered by events. GitHub repo and runnable examples exist; the implemented core works and demonstrates how to swap real infra later.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:23Inference to features overview
- 3:10Machine learning platform
- 8:04Business use case and design
- 13:24Setting up project
- 16:34Inference engine foundation
- 24:51Domain driven design
- 33:22Repository pattern and abstraction
- 40:06Model and feature repositories
- 51:15Rules-based models and features
- 1:08:56System architecture with services
- 1:13:32Event-driven design and message bus
- 1:21:34End-to-end testing
- 1:25:43Challenge questions and next steps
related talks
28mBuilding Developer-Centric ML Inference Platforms
36mBuild prod ML systems with feature /training /inference pipelines
29mBuilding an ML Platform from Scratch
18mBeyond Inference: Bringing ML into Production
43mDesigning our own in-house Feature Platform
32mTeam Teardown: Evolving an ML Platform
01h 12mBuild Simpler Production ML Systems using Feature/Training/Inference Pipelines
32mTeam Teardown: Organizing and Supporting ML