GranaÊ: AI-Powered Personal Finance Agent on WhatsApp

The Problem Personal finance apps suffer from high churn because manual entry is tedious. Users forget to log expenses, making the data useless. I wanted to remove the friction entirely by bringing the ledger to the app people use every day: WhatsApp.

The Solution GranaÊ is a frictionless finance tracker. There is no app to install. Users simply text their expenses naturally (e.g., "Dinner 45.90" or "Uber to airport 25"), and the AI handles the rest.

Zero-UI Data Entry: Leveraging GPT-4, the system extracts the amount, merchant, and date from unstructured text. It automatically assigns categories (e.g., "Food", "Transport") with high accuracy.

Natural Language Queries: Users can ask "How much did I spend on Uber this month?" or "Show me my weekly report", and the bot generates instant text summaries and charts.

Interactive Editing: A state-machine flow allows users to edit or delete transactions via numbered menus within the chat.

Hybrid Support Module: I built a separate web-based Admin Dashboard where human support agents can view ticket queues and intervene in WhatsApp conversations in real-time if the AI cannot resolve an issue.

🔄 Automating Business Logic with n8n While the core bot logic resides in Node.js, I decoupled the business operations (subscription & onboarding) using n8n workflows to ensure reliability and omnichannel communication.

The Workflow Above (Subscription Activation): The screenshot demonstrates the automated flow triggered immediately after a user subscribes:

Key Generation: A Code Node generates a unique activation key and calculates the expiration date.

Database Sync: It conditionally checks if the user exists in Supabase and updates/creates the record with the new license details.

Omnichannel Delivery: The flow simultaneously sends the activation key via WhatsApp Cloud API and adds the user to an onboarding list in ActiveCampaign (Email), ensuring 100% deliverability.

Methodology: Rapid Vibe-Coding This project showcases the speed of AI-assisted development. I utilized LLMs to generate the Sequelize models, handle the boilerplate for WhatsApp Webhooks, and optimize the SQL queries for the reports. This allowed me to focus on the high-level architecture and the user experience (UX), delivering a production-ready bot in record time.