Metarank
lastmod 2025-10-16created 2025-10-162 edits
Real-time personalization service
About
- An open-source ranking service that personalizes semantic/neural search results in real time and helps build recommender systems
- Supports Learning-to-Rank reranking on top of existing search engines, LLM-based semantic search, collaborative filtering recommendations, and more
- Integrates customer signals such as clicks and purchases into ranking to provide CTR optimization, and tracks session/visitor profiles for real-time personalization
- Fast reranking with latency around 10–20ms, and supports horizontal scaling based on Redis state management to handle thousands of RPS
- Computes dozens of common ranking signals such as CTR, Referer, UA, and time on the fly, and simplifies log ingestion through data source integrations
- Can be started in standalone mode within 1 minute via Docker, and data import, ML model training, and API startup can be done with a single command
- What can you build with it?
- Semantic Search: Elasticsearch/OpenSearch understands the meaning of queries using the latest LLMs
- Recommendations: build collaborative filtering (ALS) and content-based/semantic recommender systems
- Learning-to-Rank: fine-grained optimization of existing search, integrated with batch/online training