This AI system answers questions using a fixed set of 84 FinanceBench source documents, covering 10-K filings, 10-Q filings, 8-K reports, and earnings-call transcripts. In internal evaluation on 150 benchmark questions, the system produced correct answers in approximately 76% of cases. You may review the full document list before asking questions.
Responses are generated automatically and may occasionally be incomplete or incorrect. Please verify critical information against the original filings.
This project is a scaled-down prototype of a retrieval-and-reasoning pipeline that can be expanded to handle much larger financial document collections with minimal architectural changes.
This system is under active development, and reliability is continuously being improved.
Thank you for trying it out.