What it does
DocuMind is a retrieval-augmented assistant that only speaks from the documents it was given. Every answer carries numbered citations down to clause and page, any sentence without a source is dropped before it reaches the user, and a question the documents cannot answer gets a refusal rather than a guess. Contract, maintenance and support teams use it to find the one paragraph that matters in seconds.
What you get
- The ingestion pipeline for PDFs, spreadsheets, Office files and scans, keeping page and section references on every chunk.
- The retrieval stack — dense multilingual embeddings, BM25 for exact terms such as clause numbers, and a cross-encoder re-ranker.
- Generation with an open-weight instruction model served on one GPU, with the retrieval trace beside every answer for review.
- Web UI, API, single sign-on, an evaluation harness, an onboarding session and a direct line to the engineers.
Deployment and licensing
The one-time licence is perpetual and runs on your own hardware or private cloud — one GPU with 24 GB of memory serves a department — and includes twelve months of updates. The subscription is hosted by us in an EU data centre with a dedicated instance per customer and can be cancelled monthly. Both keep every document inside the tenant boundary.
Proven in production
On an evaluation set of 1,912 questions written by a contract team the assistant scores 0.94 faithfulness and refuses 96 % of the questions the documents cannot answer, at a median response time of 1.5 seconds. The case study linked above has the details.