Late-Interaction Vector Database
Delivering better retrieval results for reliable Gen AI
Delivering better retrieval results for reliable Gen AI
Index and retrieve the most relevant documents based on token embeddings.
Streamline development with our embedded database. Remove operational barriers to delivering on production quality.
Model | MRR@10 |
---|---|
DocT5 | 27.6 |
ANCE | 33.0 |
SPLADEv2 | 36.8 |
coCondenser | 38.2 |
RocketQAv2 | 38.8 |
ColBERTv2 | 39.7 |
MS Marco dev set scores taken from: ColBERTv2 paper
ColBERTv2 consistently has the highest relevancy. It's been used to build some of the most advanced Question Answering systems.
Reduce hallucination by incorporating late interaction into your system. LintDB is one of the only vector databases that supports it.
LintDB is the only database that supports late interaction. Increase your search relevancy for improved Gen AI performance.
Simplify Gen AI development with our embedded database. Go to production without the operational concerns.
Use state of the art techniques to drive better relevance.
Support multiple tenants from one database.
LintDB supports merging indexes and read-only replicas. Serve one index from multiple processes.
LintDB queries your data in milliseconds. Get high performance with top notch retrieval.
LintDB is an open source project powered by individual contributions.