🤓 Latest Submissions
Gbam: a record book you talk to
I run a real business and keep its books in my own record book, PriceKeeper. In 5 months I've recorded over 8 million naira of sales there. The slow part is always the same: a customer is waiting, my hands are full, and I still have to remember who paid and who owes. Gbam is the voice layer I built on top, so I can just say what happened. WHAT IT DOES I say "Two bags of rice, three thousand each, to Musa" or "Chinedu don pay 5k". English and Pidgin both work. Gbam reads the sale back with the customer's new balance, and nothing is saved until I say yes. The app does all the maths. The AI never calculates money. HOW ASSEMBLYAI IS USED - Voice Agent API: hears me, decides what to do and speaks back. - Key terms: my shop's customer names and products go to AssemblyAI. On the proof page (/proof.html) the same voice is heard twice: you can hear the same voice with and without key terms and compare. The difference shows most on harder names. - Streaming speaker labels (Universal-3.5 Pro): haggle mode tells me and the customer apart. ALSO IN THE APP - Daily close: say "close the day" for a spoken debrief and a summary to send on WhatsApp. - Memory: remembers who we were talking about, never amounts. Browser only, 12 hours, with a Forget button. - Haggle mode (early): picks the last price said and prepares the sale. Tested on recordings only. - WhatsApp receipts, low-stock warnings, undo by voice, a wrong-price guard, and a dashboard and trades list built from the same records. TRY IT https://gbam-chi.vercel.app (no login; a sample shop saved only in your browser). Code: https://github.com/Sireadell/gbam LIMITS English and Pidgin only. AssemblyAI doesn't support Hausa, Yoruba or Igbo for live speech yet. Haggle mode is early.
30 Sep 2026
