daftari
For many small businesses and households, the financial record is a WhatsApp chat full of forwarded M-PESA messages.
A tool that reads an exported WhatsApp chat and turns the transactions in it into clean records you can filter and export.
- Unstructured Data Extraction
- AI-Assisted Classification
- Human Review
- Practical Automation
The problem
It started as a problem of my own. The payment messages were all there in the chat. Nobody could total them, and nobody fancied retyping them either.
What I built
The parsing is done by rules. Some four hundred lines of patterns recognise the message formats used by mobile money and the banks, and AI only suggests a category. Every transaction gets a confidence score, and the uncertain ones are set aside for a person to check.
What it does
- Upload a WhatsApp export and get back a table of dated, categorised transactions.
- A review filter that gathers every low-confidence record in one place.
- Search and filters by category or merchant.
- Charts showing where the money went.
- Export to Excel, or to CSV and JSON.
Under the hood
- Next.js and TypeScript.
- Rule-based parsing first, then Gemini categorisation checked against a fixed schema.
- Rate limiting on the upload endpoint.
Where it stands
It is live and in use. It is also the clearest example of what I like doing for a business, which is taking the records it already keeps and making them countable.
Why it might matter to you
If your organisation’s real records live in chats or spreadsheets, this is a sensible place to start.
Next Case Study
Golden Eagle Insurance Agency
An insurance agency trading since 2006 needed a website that did more than list its products.