AWS Bedrock · Python · SQL · React

AI Finance Data Agent

A conversational agent for financial analysis: a question asked in plain English returns a written answer together with the chart and the underlying numbers it was drawn from. Built on AWS Bedrock over a SQL data layer, with suggested follow-ups that carry an analysis past its first answer.

Natural language queries
Answers with evidence
Guided follow-ups

What it demonstrates

Plain-language questions against a data model: Questions are resolved against the underlying tables, so an answer is retrieved from the data rather than recalled from whatever the language model already believed about it.
Every answer shows its evidence: The written conclusion arrives with the revenue trend and the product-level breakdown beside it, so a reader can check the claim against the numbers instead of taking it on trust.
Takeaways separated from narrative: The points worth carrying into a meeting are pulled out on their own, which is the part a written summary usually buries in prose.
Follow-ups the agent proposes: Suggested next questions, like comparing to last year or breaking down by region, keep an analysis moving instead of ending wherever the first question happened to stop.

Reproduction

The interface

The agent answering what drove revenue growth this quarter: a written answer, a revenue trend chart reading $2.48M, a revenue-by-product table, and three key takeaways — Asking what drove revenue growth returns the written answer, the trend behind it, and the product-level numbers it was drawn from, with the next questions worth asking suggested rather than left to the reader.

The agent comparing Q1 and Q4 profitability: net profit, net margin, gross margin and operating margin tiles, a margin trend chart, and a table attributing the change to each driver — Asking what drove revenue growth returns the written answer, the trend behind it, and the product-level numbers it was drawn from, with the next questions worth asking suggested rather than left to the reader.