What it does TensorRec is a Python framework that wraps TensorFlow to train and score recommendation models. You feed it user_features, item_features, and interactions; it handles the data plumbing, scoring, and ranking. The hook is that you can swap in your own embedding functions and loss functions without rewriting the boilerplate.

The interesting bit The project is explicitly dead — the README’s second section is a retirement notice. The author still accepts pull requests but won’t push planned updates, and helpfully points to three successors (TensorFlow Ranking, Spotlight, LightFM). It’s a rare clean exit in open-source ML tooling.

Key highlights

• Customizable representation and loss functions via TensorFlow
• Built-in evaluation metrics like recall@k
• Generates dummy data for quick prototyping
• pip-installable (pip install tensorrec)
• ~1,300 GitHub stars; last meaningful activity unclear from README

Caveats

• Not under active development — explicitly stated by the author
• README offers no benchmarks, performance numbers, or production usage claims
• Unclear which TensorFlow versions are supported; no compatibility matrix visible

Verdict Worth a skim if you’re studying recommender system architecture or maintaining a legacy codebase that already uses it. Everyone else should start with the author’s own suggested alternatives — particularly TensorFlow Ranking or LightFM — rather than adopting a frozen framework.