Show HN: AuraSpend " Voice-first expense tracker using Gemini for NLU
subhanzg Saturday, February 14, 2026
I built an expense tracker where the primary input is voice. The thesis: expense tracking apps fail because manual entry has too much friction, which is why most people abandon them within weeks.
How it works: - Open app, tap mic - Say "yesterday I spent 30 dollars on lunch at Chipotle" - AI extracts: amount, merchant, category, date - One tap to confirm and save
Other features: - Receipt scanning with ML Kit OCR + Gemini Vision for parsing - Background notification listener that catches bank alerts and creates pending entries - GPS-based currency detection for travelers - Multi-language with native script fonts (Nastaliq for Urdu, Devanagari for Hindi, Amiri for Arabic) - Offline-first with Google Drive sync
Tech: Flutter, Riverpod, Hive, Gemini 2.0 Flash
The hardest part was prompt engineering for consistent JSON output across languages. Had to build "Supreme Laws" into the system prompt - things like "never translate native text" and "extract currency from context."
Free tier available. First 500 users get Pro free for a year.
Play Store: AuraSpend
Happy to discuss architecture, prompt engineering approach, or take technical feedback.