Skip to main content

AI Financial Statement Analyzer

  • Python
  • OpenAI API
  • Streamlit
  • Pandas
  • ai-tool
  • in-progress

Status: Prototype

Evidence highlights

  • Cuts first-pass commentary time materially on clean sample exports versus manual copy-paste workflows.
  • Improves governance posture by generating narrative only from computed metrics, not raw statement text.
Request walkthrough (opens in new tab)

Purpose

This project tests the downstream half of a reporting loop: once metrics exist, how quickly can a team reach a defensible first-draft narrative? The prototype accepts structured financial exports, computes a standard ratio set in Pandas, and sends the metrics—not raw statements—to an LLM for a short summary.

Design Iteration

An early approach pasted more of the source file into the prompt. That was faster to prototype and harder to trust: models can smooth over missing lines or imply trends that are not in the data. The revised approach uses structured metrics only and treats narrative as commentary on computed numbers, not a substitute for them.

Current Status

The prototype supports end-to-end flow on clean CSV-style inputs: upload, review ratios, edit the draft summary. On sample three-statement exports, first-pass commentary time can drop materially versus manual copy-paste—subject to review and edits.

Still open: broader statement formats, configurable ratio libraries, authentication for team deployment, and retention settings aligned with enterprise LLM policy.

Tech Stack

  • Python — Core logic and data processing
  • OpenAI API — Narrative from structured metrics (provider swappable)
  • Streamlit — UI for quick iteration
  • Pandas — Loading, ratios, trend flags
  • Wire governance earlier: log which metrics entered each prompt, block prohibited upload patterns, and export a review trail for audit.
  • Separate “explore in Streamlit” from “run behind SSO”—demo and production paths diverge quickly.
  • Code: Private while core prompt and validation logic are still being hardened.
  • Evidence: Product walkthrough and sample output can be shared on request via the contact page.

Related: Automated monthly reporting pipeline for upstream data patterns, and AI governance for finance teams before productionizing LLM output.