Inspiration

I kept watching small business owners around me — market traders, shop owners — struggling with something that should be simple: keeping track of their own money. Sales scribbled in a notebook if recorded at all, no real sense of profit or loss, stock counted from memory. I wanted to build something that could actually help, without asking them to learn a new app or change how they already work. That's what TradePulse AI started as: a rescue mission for people who'd never had a proper ledger before.

What it does

As I built it, I realized the idea wasn't just for struggling micro-traders — any business, big or small, that values convenience could use this. You send a text, a voice note, or a photo of a receipt to a Telegram bot, and it turns into a real, categorized ledger entry: income vs. expense, quantity and unit price separated out, stock automatically adjusted when something sells or runs out, even a suggested customs code for cross-border goods. No forms, no training.

How I built it

It's a Flask app on PythonAnywhere, backed by SQLite, talking to Google's Gemini API for every intelligent decision — parsing free text and photos into structured data, transcribing voice notes, screening subscription payment proof before a human approves it, and answering business questions grounded in the trader's real numbers. Telegram is the entire interface, since it's already installed on the phones people actually have.

Challenges I faced

This is the first app or bot I've ever built, end to end — that alone was both thrilling and genuinely hard. I hit real production problems: Gemini models getting deprecated or access-restricted mid-build, silent API failures that I had to learn to catch instead of letting them corrupt data, quota limits I didn't know existed until my bot stopped responding. Debugging a live system for the first time, reading stack traces, understanding webhooks — none of it was familiar going in. But Gemini itself became the tool that helped me reason through fixing the very system it powers, which was a strange and satisfying loop.

What I learned

Beyond the technical skills, I learned how much thought a "simple" bot actually requires once real money and real inventory are on the line — every shortcut has a consequence for someone's business records. I'm a BSc Computer Science and Mathematics student in my third year at the University of Eswatini, and this was my first real proof that I could take an idea from a problem I saw around me all the way to something running in production.

What's next for tradepulse-ai

If the consumers are very happy with the services the bot provide and the demand increase , I plan to hire people that can help me build the actual app of the bot and help in maintenance of smooth operation of the app.

Built With

Share this project:

Updates