itsdersty/django-neural-feed

Django Neural Feed (DNF) is a Python library that adds personalized content recommendation feeds to Django applications. It works by storing vector embeddings of content items in PostgreSQL using the pgvector extension, which allows similarity calculations to happen inside the database rather than in application code. When a user interacts with content, the library builds a vector profile of that user's preferences, then scores new content items by combining similarity to that profile with how recent the content is and how popular it is, all in a single database query. The library is installed as a Django package. You configure it by creating a feed class that specifies which content model to track, which interaction model to monitor (such as a likes table), and how to weight similarity against freshness and popularity. The library attaches to your models through Django signals, so you do not need to modify your existing interaction tables. User preference profiles are stored separately from the main user model, one profile per feed type. This means you can run multiple feeds with different configurations for different parts of your application without them interfering with each other. Background work such as generating embeddings and aggregating user vectors can run through Celery for async processing, with an automatic fallback to synchronous background threads if no Celery broker is running. The default encoder for generating embeddings uses sentence-transformers, which is an optional install since it pulls in PyTorch. If you have a custom encoder such as OpenAI or a hosted API, you can skip the local model entirely. The README includes detailed code examples for all configuration steps, including model setup, feed class definition, and querying for personalized results. This library is aimed at Django developers building content-heavy applications where personalized ordering matters, such as social platforms, news feeds, or content aggregators.