Budget ON
A personal budget management with AI receipt scanning, dynamic daily spending limits, and shareable monthly reports designed for WhatsApp.
Type
Solo Project
Timeline Start
Sep 2026
Status
Beta Release
Platform
Android Native
Motivation
Why I Built It
Most budgeting apps fail because the cost of logging is higher than the value of the insight. Manual entry creates psychological friction that leads to abandonment within weeks. Budget ON was engineered around a single constraint: frictionless data capture. By combining Gemini Vision API for zero-typing receipt parsing with WhatsApp-native reporting, it aligns with how Indonesian users actually manage and share financial information.
Core Capabilities
Key Features
01
Multi-Modal AI Receipt Scanning
Capture receipts via camera or gallery. Gemini Vision API parses messy thermal paper into structured transactions with strict JSON validation, rejecting hallucinations before they reach the UI. Includes a 90s timeout and manual fallback for reliability.
02
Dynamic Daily Spending Guardrails
Set personalized daily spending ceilings. The app tracks real-time expenditure against this limit and triggers high-priority native notifications the moment you overshoot, acting as an immediate behavioral nudge rather than a retrospective report.
03
WhatsApp-Native Monthly Reporting
Generate consolidated monthly summaries formatted specifically for Indonesian sharing habits. One-tap export sends plain-text breakdowns directly to WhatsApp chats, eliminating friction in communicating financial status with partners or family.
Architecture
Engineering Decisions
Traditional on-device OCR libraries like Tesseract consistently failed on faded thermal ink, skewed angles, and unconventional Indonesian receipt layouts. Gemini's multi-modal reasoning handles these edge cases with high accuracy. The trade-off is dependency on network availability and API quotas, which is mitigated by a robust offline manual-entry fallback.
What's Next
Future Update
Future development focuses on replacing the current 1,000-document transaction cap with date-windowed pagination to support long-term users without performance degradation. I am also exploring alternative vision models or lightweight local OCR fallbacks to reduce dependency on external API quotas during peak demand. Next feature priorities will be determined by gathering direct user feedback through the in-app form to ensure development aligns with actual usage patterns rather than assumptions.