I Built a Local AI Finance Coach for My Friend's Bubble-Tea Shop - No Cloud, No API Keys

Description and use

A local AI finance coach processes bank statements on the owner’s machine, producing plain‑language reports without cloud exposure.

Problem

A small business owner needed a way to understand her monthly bank statement without uploading sensitive data to external services.

Outcome

The tool produced a concise report confirming profitability, identifying key income and expense drivers, and recommending supplier negotiation and rent review.

Workflow

• Upload Excel/CSV file to local analyzer.py
• Analyzer auto‑detects date, description, income, expense columns using fuzzy matching
• Compute monthly income, expenses, net flow, top sources and categories, biggest transactions
• Output compact JSON summary to reporter.py
• Reporter invokes local llama.cpp with Qwen2.5‑3B‑Instruct to generate a 250‑350 word business report
• Display report and tables in Gradio UI

Evidence

Bank statements are the last thing you want on a stranger's server. Here, the file, the statistics, and the model weights all live in one local process.

Source

Feed it three months of a bubble‑tea shop's transactions and it produces a report like this...

Source