Automatic data extraction from PDF and DOCX, contract and act verification, contractor price control, AI review of requirement specs against standards. On-premise.
FAQ
What is AI document control?
An AI system that parses corporate documents (contracts, acts, estimates, invoices, requirement specs, insurance programmes) and applies business logic: it finds violations, overpayments, deviations from the contract and breaches of standards or the compliance taxonomy. The output is structured data linked back to the source fragment, a queue of cases and finished documents (claims, reports, spreadsheets).
How do you guard against LLM hallucinations?
Three layers. First, every output must carry a quote from the document, and a fuzzy match against the original catches invented references. Second, a pass-2 NLI validation for suspicious outputs. Third, sorting by confidence: auto-approved / spot-check / requires-review — the expert only looks at what is doubtful.
Where does the data live? Is on-premise possible?
Yes, on-premise inside the customer perimeter is the main deployment mode. The stack is Keycloak SSO, PostgreSQL 18, Qdrant, MinIO and a LiteLLM gateway. Data never leaves the perimeter.
Is it a boxed product or a platform to customise?
A universal core plus scenario configurations. Five scenarios run in production: price control, obligation control, contractor history, AI review of requirement specs, compliance matrix. A new scenario is a new configuration — rules, prompts, taxonomy profile — with no code fork.
How is ROI calculated?
Each scenario has its own metric. Price consistency — money saved against the median. Obligation control — money recovered and recovery rate. AI review and the compliance matrix — expert hours removed (hours to minutes) and missed violations found after deployment.