Bye Sub

Bye Sub is an AI co-pilot for your bank account, running on GKE. It hunts down forgotten "money vampire" subscriptions in your transaction data and empowers you to cancel them, saving you real money.

Comment

Inspiration

How many of you have lost money on a subscription you forgot about? Hidden fees are like money vampires — draining your wallet silently. I wanted to build an AI-powered ghostbuster that saves users from sneaky charges.

What it does

Bye Sub enhances the Bank of Anthos experience with a new, intelligent microservice deployed on Google Kubernetes Engine. My service analyzes raw transaction data, uses a hybrid model to detect recurring subscription patterns, and leverages the Gemini API to generate clear, actionable guides on how to cancel each one. It turns confusing bank statements into a simple tool for financial control.

How I built it

I built Bye Sub as a new, containerized microservice deployed on GKE. The architecture consists of:

  • A Python/FastAPI Backend: This is the core service running in a Docker container on GKE. It exposes a simple REST API.

  • Hybrid Detection Model: The service first applies smart heuristics (recurrence, known merchants, keywords) to find potential subscriptions.

  • Gemini-powered Agent: These candidates are then passed to a Gemini agent, which verifies them and generates human-friendly cancellation instructions.

  • Bank of Anthos Adapter: To ensure compatibility with the real API, I analyzed its source code, specifically the Transaction.java model. Based on this, I built an Adapter that transforms BoA's raw transaction data into the format service needs. The demo uses mock data in this exact BoA format.

  • A React/Vite Frontend: A clean, responsive UI also deployed as a separate container on GKE.

Challenges I ran into

  • Accurately identifying subscription patterns from raw, unlabeled transaction data without merchant names.

  • Crafting effective Gemini prompts that produce concise, useful, and empathetic cancellation tips.

  • Designing a decoupled architecture with an Adapter Pattern to ensure our new service could integrate with Bank of Anthos without modifying it.

  • Managing a full-stack deployment on GKE, including Docker containerization, Kubernetes secrets for API keys, and networking between services.

Accomplishments that I'm proud of

  • Successfully designing and deploying a complete, multi-container application (frontend & backend) on GKE.

  • Implementing a robust Adapter Pattern, proving service's compatibility with the Bank of Anthos API contract.

  • Creating a powerful hybrid detection model that combines predictable heuristics with the generative power of Gemini.

  • Building a fun, engaging, and useful product that solves a real-world problem.

What I learned

  • The power of a hybrid approach: combining deterministic algorithms with Gemini yields more accurate and reliable results.

  • How to design for integration: the Adapter Pattern is crucial for building new microservices that connect to existing systems like Bank of Anthos.

  • The complete lifecycle of a cloud-native application: from local coding to Docker containerization and live deployment on GKE.

What's next for Bye Sub

  • Add real-time monitoring of transactions with instant notifications

  • Expand detection to include other “money leaks” (unused SaaS, forgotten trials, duplicate services)

  • Provide one-click cancellation flows via partner APIs

  • Deploy Bye Sub as a plug-in for digital banks and fintech apps worldwide

Share this project:

Updates