Case · AI & documents

AI that reads your documents, and shows its sources

ReportAnalyzer is our own RAG platform for annual reports, papers and long reports. The main design goal: no answer without a source you can check.

Type
Own R&D platform
Use
Analysing annual reports, papers, reports
What we did
Architecture, AI pipeline, evaluation, interface
Technology
RAGDoclingPostgreSQLpgvectorEvals
AI that reads your documents, and shows its sources

The challenge

General chatbots answer questions about documents fluently, but sometimes invent figures or sources. For accountants, analysts and consultants that’s useless: every answer has to be traceable to the document itself.

What we built

  • Documents are processed keeping structure, tables and page numbers
  • Answers quote verbatim, with page numbers
  • An AnswerGuard checks every quote against the source and refuses to answer if it doesn’t match
  • Automatic extraction of fixed data points, such as FTE and sustainability targets
  • Local-first: can run entirely on your own hardware, with no data going to the cloud

Result

  • Quotes checked against the source
  • Page-level references
  • Design recorded in 18 architecture decision records (ADRs)

Sound familiar?

Familiar to anyone who reads a lot and can’t afford to miss things: accountants, consultancies and engineering firms, lawyers, insurers, researchers. We connect this approach to your documents and systems, on your hardware or in the EU.

Discuss your situation →

Where is your business losing time or money right now?

Book a free 30-minute call. You get honest advice on what we would build, what it costs and what it returns, even if you don’t work with us afterwards.