Traditional full-text search looks for words, not meaning. With the help of artificial intelligence, we build search that understands a natural-language query and helps people discover things they would never have found on their own.
What institutions come to us with
- Users do not know the exact term, look for “something like…” or ask in whole sentences.
- Historical texts use period language and different words than today’s readers type.
- Extensive archives are searchable, but there are so many results that what matters gets lost.
- Recommendations and connections are missing — similar topics, related works, other documents about the same person or place.
What we focus on
- Natural-language queries. The system evaluates the intent of the query, highlights what matters in the results and handles follow-up questions.
- Discovering connections. Recommendations of similar records, grouping of different editions of one work or searching for similar motifs.
- Data enrichment. Recognition of people, places and dates in text, summaries, abstracts and vectorization that make it possible to work with the content in depth.
- A curation layer. An expert can sort results into collections and series, correct and edit them — AI helps, but does not decide for them.
How we work
- We connect the sources and metadata that the search will draw on.
- We choose suitable models for the given language and type of content and deploy semantic search and recommendations.
- We store enriched data as a separate layer. Archival packages remain untouched and new outputs can be updated, deleted or regenerated at any time.
- We measure relevance and fine-tune the results according to how people actually search.
What you get
- Results by meaning, not just by matching words.
- Faster work with extensive archives for the public and researchers alike.
- New paths to content through recommendations and connections.
- Control over your data — enrichment without changing the original.
Where it already works
Paměť novin (Newspaper Memory) makes newspapers from the turn of the 19th and 20th centuries accessible: it understands natural-language queries, can interpret source texts and greatly speeds up work with an extensive archive. We are bringing semantic search, similar image and document search, automatic translations, text-to-speech and article summarization to Kramerius. Makarius searches sheet music by a given melody and suggests similar musical works.
Transferable to: knowledge bases, documentation search and customer support.
