AI GitHub
Infinity for hybrid search in one engine
Combine vector, keyword and other search signals over the same document collection.
Key Takeaways
- Infinity supports multiple search modes through one engine.
- Use a dataset with both searchable text and embeddings.
- Prepare a test set containing both exact part numbers and paraphrased questions.
- Log the candidates from each mode so you can tell whether a failure came from retrieval or ranking.
How it works
Infinity supports multiple search modes through one engine. A query can bring different evidence together when exact names and semantic similarity each matter.
What you need
Use a dataset with both searchable text and embeddings. Define where filtering, candidate selection and score fusion happen so individual pieces can be measured.
What to try first
Prepare a test set containing both exact part numbers and paraphrased questions. Compare individual retrieval modes with the combined ranking.
Include a query whose wording matches an irrelevant document exactly. This helps expose whether hybrid ranking gives the keyword signal too much weight.
Where to look for failures
Fusion can obscure why a result moved. Log the candidates from each mode so you can tell whether a failure came from retrieval or ranking.
Related Franklin pages
Start with a labeled query set and save each engine's candidates. Compare filtered search and ranking behavior before extrapolating throughput from a query without metadata constraints.
- Build search applications that combine text, vectors and ranking functions. Vespa for custom ranking and retrieval
- Search embeddings alongside structured metadata in a dedicated database. Qdrant for filtered vector retrieval
For the role of the index in nearest-neighbor search, see Franklin’s glossary. Vector indexes
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗