A Go recommender engine that bridges collaborative filtering and LLMs

What it does

Gorse is an open-source recommendation engine written in Go. You import items, users, and interaction data, and it automatically trains models to generate recommendations exposed through RESTful APIs and a web dashboard. It supports multiple strategies—latest, user-to-user, item-to-item, and collaborative filtering—and stores data in MySQL, Postgres, MongoDB, or ClickHouse.

The interesting bit

Rather than forcing you to choose between classical matrix factorization and newer neural approaches, Gorse supports both classical recommenders and LLM-based rankers, plus multimodal embeddings for text, images, and video. Its architecture splits duties across a single master node for training, worker nodes for offline recommendations, and server nodes for real-time API traffic.

Key highlights

• Supports classical strategies—latest, user-to-user, item-to-item, and collaborative filtering—alongside LLM-based rankers.
• Handles multimodal content such as text, images, and video via embeddings.
• Single-node training with distributed prediction: a master node trains models, worker nodes generate offline recommendations, and server nodes handle real-time APIs.
• Includes a GUI dashboard for pipeline editing, monitoring, and data management.
• Backs into MySQL, MariaDB, Postgres, MongoDB, or ClickHouse, with intermediate caching in Redis or the database layer.

Caveats

• Model training is confined to a single master node; distributed scaling applies only to prediction.
• The README advertises LLM-based rankers but provides no specifics on supported models, latency, or cost.

Verdict

Consider this if you need a Go-based recommender with a built-in dashboard, pluggable storage, and a mix of classical and modern ranking strategies. Pass if you require distributed model training or deep operational control over LLM inference.