How triiju/hybrid-recommendation-engine works
Project type and purpose
• This is a full-stack, modular hybrid recommendation engine written in Python.
• It provides both an offline batch pipeline (data download, preprocessing, baseline models, tuning, evaluation, cold-start experiments) and an online serving layer (FastAPI REST API).
File structure
Top-level directories:
• api/ – FastAPI application for serving recommendations and metadata
• scripts/ – Standalone automation scripts (alpha sweep, end-to-end pipeline)
• src/ – Core Python modules implementing data loading, preprocessing, recommenders, evaluation, logging, hyperparameter tuning, and the API entry point via uvicorn
• models/ – Serialized model artifacts (e.g., neural_cf.pt)
• experiments/ – Empty placeholder for experiment outputs
• tests/ – Unit tests for key modules and metrics
Configuration and build files:
• Makefile – Orchestrates reproducible pipeline targets
• requirements.txt – Python dependency list
• vercel.json – Deployment configuration (likely for frontend or serverless host)
README for additional insights
• Offers a clear breakdown of each component’s file, function, and purpose
• Defines the offline workflow (download → preprocess → baselines → neural → hybrid → evaluate → alpha sweep → cold-start)
• Exposes a FastAPI-based REST API with three endpoints: /recommend, /health, /meta
• Mentions external dependencies: MovieLens 100K dataset, PyTorch, Optuna, FastAPI, uvicorn