Problem
Traditional expense tracking tools are often rigid and time-consuming — every transaction requires manual input, category selection, and validation. As a result, users (myself included) quickly lose consistency, leaving analytics incomplete and insights unreliable.
Solution
I built a proof-of-concept automation that turns WhatsApp into a personal expense assistant. Behind the scenes, the system listens for messages, parses text or images using Gemini API, and pushes structured data to Google Sheets. Everything runs locally on a Dockerized home lab server (Ryzen 5, 8GB RAM) monitored through Grafana (cAdvisor) for lightweight observability.
Technical Highlights
- WhatsApp automation using
wweb.jsfor real-time message ingestion - AI-powered text & OCR extraction via Gemini API
- Google Sheets integration for persistent structured data
- Dockerized microservice setup with full monitoring stack
- Scalable flow design, ready for multi-user deployment
Next Steps
- Introduce weekly/monthly report summaries and budget reminders
- Explore privacy-focused open-source LLMs (Mistral, Qwen) for local inference
- Refine data pipeline and add smart tagging for recurring transactions
Reflection
What started as a small daily frustration evolved into a working prototype of an AI-driven automation tool. This project reminds me that most meaningful innovations begin with small personal annoyances — you just have to build through them.
Stay curious. Stay foolish.
Gallery
