Tessera was born from a simple observation: organizations sit on massive document corpora — contracts, filings, agreements, records — and the intelligence locked inside them is invisible. Not because search doesn't work, but because the metadata required for intelligent search doesn't exist yet.
Traditional document AI answers the question you asked. Tessera discovers the questions you should have asked. It does this through emergent metadata — an AI-driven process that analyzes an entire corpus and discovers the thematic schema latent in the documents, without being told what to look for.
The name comes from the individual tiles in a mosaic. One tile means nothing. Assemble thousands and a picture emerges that no single tile could reveal. That's what Tessera does with data: each document is a tile, and the platform reveals the mosaic.
“One document has metadata. A thousand documents have a pattern. A million documents have intelligence.”
We believe the barrier to document intelligence is not better models or faster search — it's that the metadata required for thematic queries doesn't exist until someone creates it. We believe AI can discover that metadata without being told what to look for. And we believe the proof is in the findings: Tessera has analyzed over 2,400 contracts across two completely different domains, with zero configuration change, and produced findings that domain experts recognize as genuine and actionable.
We also believe in transparency. Tessera tells you when its confidence is low. It shows you the evidence behind every claim. It won't fabricate certainty where evidence is thin. We call this the dark hallway principle: better to admit where the light doesn't reach than to pretend the whole building is illuminated.
Tessera is built on open-source infrastructure wherever possible. Our vector search layer runs on Vespa, the open-source AI search platform trusted by Spotify, Perplexity, and Yahoo at billion-record scale. Our knowledge graph runs on Neo4j. Our embedding models are open-source. We believe your data should be portable and your infrastructure should be auditable.
Interested in running Tessera on your own corpus? Have a dataset that would make an interesting case study? We'd like to hear from you.
hello@tessera.nowEvery record is a tile. See the whole picture.