Schema-driven extraction across millions of documents, with provenance on every field.
The result is a knowledge graph of relationships across your entire corpus.
Run a pilotLissajous is an inference platform for document repositories at scale. You define an extraction schema that's filled document by document with provenance and confidence scores on every field. The results are then assembled into a knowledge graph of entities and the relationships between them.
At that scale, relationships surface that nobody could find by hand, and the graph becomes a backbone you can build search, analysis, and applications on.
Profile a repository before you spend anything on it. Documents cluster by type and the platform proposes sets to run, so you pick from what's there instead of writing filter rules.
Millions of documents, processed in batches, running against your data where it already lives.
The schema determines the model. Each page or field goes to the right one, whether that's bespoke, open, or frontier.
Bring your own keys and own the spend. Virtual per-project keys give you chargeback and hard caps, and jobs are batched and scheduled against those limits.
An agent drafts the schema from your own documents. Your domain experts correct it rather than building from nothing.
Provenance on every field and a confidence score beside it. Your corrections recalibrate the scoring over time.
Runs in your cloud, our cloud, or on-prem. The platform deploys where your data already lives, and nothing is exported to us.
Connects to object stores, document management systems, and data lakes. Documents are read in place and never copied. The extracted data and knowledge graph are stored in your environment, alongside your own systems.
Frontier models, open weights, local models, or your own fine-tunes. The registry advertises cost and capability, and the platform optimizes model selection per schema field — the right one for each kind of extraction. Pin a model to any field or section when you need to.
We integrate the platform into your environment — data sources connected, models registered, outputs wired into the systems that consume them. You get something running, not a deployment to figure out.
Tell us what you're working with and we'll set up a call.
