কত বাকি (Koto Baki)
Small Bangladeshi shop owners track credit sales (baki) by hand in paper ledgers; typing transactions into a digital system is friction they won't adopt.
Technology Architecture & Stack
Designed a voice-first digital khata where shopkeepers speak transaction details naturally in Bangla—whether recording cash sales, issuing credit, or logging repayments. Gemma 4 native audio processing directly ingests and parses raw speech into structured ledger entries without requiring separate STT steps. The UI mimics a traditional ruled ledger notebook complete with physical page-turn animations via GSAP and React Pageflip. Low-confidence speech parsing triggers visual user confirmation prompts before writing to SQLite.
Key Features & Functionality
- Native audio parsing with Gemma 4 for natural Bangla speech processing.
- Physical ledger user experience featuring tactile page-turn animations using GSAP & React Pageflip.
- Zero typing needed: speech directly updates cash, credit (baki), and repayment ledgers.
- Low-confidence parsing verification dialogs to prevent incorrect financial records.
- Lightweight Python FastAPI backend paired with SQLite persistence.
System Architecture & Technical Design
Direct native audio processing bypasses traditional Speech-to-Text latency and error cascades, handling accent variations and colloquial Bangla context natively.
Key Engineering Takeaways
For low-tech adoption, matching real-world physical metaphors (ruled paper notebook) yields significantly higher user engagement than corporate dashboard layouts.
Project Milestone / Status
Built for the Gemma @ Bangladesh Hybrid Hackathon '26.