Service
Ask questions over your records, filings and contracts. Every answer cites the passage it came from, and the system says so when the documents do not support an answer.
The answers your team needs are in documents: a chart, a credit agreement, a tariff filing, a policy manual. Finding them means reading the same pages again for every question, and the person who knows where to look is always busy.
A general chatbot makes this worse in a way that is hard to see. It answers fluently whether or not the document says what it claims, so every answer has to be checked against the source, which is the work you wanted to save.
What saves time is an assistant whose answers can be checked in seconds: each claim tied to a passage, and a clear refusal when the passage does not exist.
Each sentence carries a link to the passage or entry it came from. A reviewer clicks and sees the source.
When retrieval does not support an answer, the assistant says so instead of guessing.
Cited passages are checked against the claim before the answer is shown; failed citations are removed and flagged.
Answers that feed a decision go to a review queue, with approve, correct and comment.
Documents are parsed by their own structure (sections, entries, clauses) so a passage arrives with its context.
Question, sources retrieved, sources cited, model and time, kept for your records.
Chart Explorer, the healthcare instance on synthetic patient records, opens at chart-explorer.df1.io when it ships.
Grounded Document Assistant
LiveAnswers questions over statutes and contracts with the source passage for every claim, and declines when the text does not support an answer.
Answers are generated only from passages retrieved from your documents, and the passage ids travel with the text.
A second pass checks that each cited passage supports its sentence. Unsupported claims are dropped, not reworded.
A fixed question set, including questions the documents cannot answer, runs on every change; the refusals are tested as carefully as the answers.
Where an error is expensive, a person approves before anything leaves the system.
Operate for one document set; Scale for several, or when the assistant feeds other workflows. Enterprise when the documents hold PHI or must stay in your cloud.
Useful add-ons: Compliance package for PHI work; Team training workshop.
Text documents with a structure: clinical charts, contracts, filings, policies, manuals, regulations. Scanned documents need OCR first, which is part of the Discovery Sprint's audit.
No. DF1 uses model providers under terms that exclude training on your data, and the documents stay in the storage you choose.
The assistant says the documents do not contain it. That answer is tested on every change, because a confident wrong answer costs more than no answer.
Production work on PHI runs under an Enterprise agreement with a business associate agreement chain covering DF1, the model provider and the host. Demos and Discovery work use synthetic or de-identified data.
Describe the process and the data. You get a reply within one business day, and a call if it fits.