Team So-low

In the Philippines, more than 1.3 million neighborhood sari-sari stores still use handwritten credit notebooks, known locally as a "listahan ng utang," to track what their customers owe. Plenty of point-of-sale and bookkeeping apps already exist, but store owners rarely stick with them. Typing items onto a phone screen during a busy rush at the counter is simply too slow compared to grabbing a pen and paper. Still, relying on physical notebooks causes real financial problems. Ink smudges, quick mental math goes wrong, and forgotten debts slowly eat away at a store's working capital. I built Talaan, which translates to "The Notebook, Kept," to fix this without forcing owners to change how they work. Store owners keep writing on paper during busy hours, and later take a single photo with their phone to turn that page into an accurate digital ledger. Talaan runs on Meta Muse Glimmer 30B (meta/muse-glimmer-30b) to read messy Taglish shorthand, item multipliers like quantity times unit price, and crossed-out lines, sorting them into credit, payment, and cash-sale entries. Because a financial app cannot afford to have an AI guess at math, I built a strict Two-Phase Bookkeeping Engine. Standard TypeScript code calculates every line item and page total down to the exact centavo. If the owner's handwritten math does not match the actual numbers, the app flags the error in a red margin on the screen so the owner can fix it with one tap. If a photo is too blurry to read, the system rejects it right away and asks for a retake, making sure that bad numbers never make it into the database. I built the app using Next.js, TypeScript, Tailwind CSS, Zod, IndexedDB for offline queueing, and Supabase PostgreSQL with Row-Level Security. Beyond reading notebook pages, Talaan helps store owners collect money and restock their shelves. It creates itemized digital receipts (E-Resibo), writes polite Taglish text reminders for unpaid balances, builds a market restock checklist (Pamalengke), and includes a store assistant ("Tanungin si Talaan") that prevents made-up numbers. Because the model runs at roughly $0.002 per scan, I can offer the app for ₱99 a month (about $1.70), a price that small neighborhood stores can actually afford.