galafis/feature-store-architecture
Feature Store para ML: armazenamento online (Redis) e offline (Parquet), API REST Flask, computacao de features com validacao e geradores de dados de exemplo.
What's novel
Feature Store para ML: armazenamento online (Redis) e offline (Parquet), API REST Flask, computacao de features com validacao e geradores de dados de exemplo.
Code Analysis
12 files read · 4 roundsA simplified in-memory feature store with Redis (online) and Parquet (offline) dual storage, a Flask REST API, and synthetic data generators for e-commerce and finance domains, intended to prevent training-serving skew in ML pipelines.
Strengths
Clean dataclass-based domain modeling with clear separation of Feature, FeatureGroup, and FeatureStore concepts; the synthetic data generators use realistic statistical distributions and the test suite covers a reasonable range of API endpoints and edge cases.
Weaknesses
Critical test bug (MockRedis implements hmset but code calls hset, meaning tests would crash), the offline store writes a new Parquet file per entity making it unusable at scale, and the Flask API is duplicated across two files with overlapping endpoints.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
23
Contributors
2
Files
22
Active weeks
4
Repository
Language
Python
Stars
1
Forks
0
License
MIT