Kaeru (カエル)

Inspiration

Canceling unwanted subscriptions or disputing charges is tedious: endless calls, confusing merchant flows, and hours lost. We built Kaeru (カエル) to fix that: a unified dashboard that detects suspicious transactions and executes cancellations or disputes automatically, saving time and frustration.

What it does

🧠 Detects suspicious or recurring transactions using Plaid.
🚨 Flags potential fraud with confidence levels and suggested actions.
🪄 Cancels subscriptions automatically via: Direct merchant APIs (when available) Voice agent (Vapi) that calls merchants and follows cancellation scripts ⚖️ Files disputes and tracks resolution progress. 📞 Displays real-time agent status, queue, and completion metrics in the dashboard.

How we built it

⚙️ AWS Services

AWS Amplify: Central platform for provisioning backend resources, defining data models, and generating configuration used by the app.
AppSync GraphQL API: Hosts the main data layer for transactions, detection items, and fraud artifacts, integrating AI routes via Bedrock.
Amazon Cognito: Handles secure user authentication with email sign-in and MFA.
AWS Bedrock: Powers AI inference for fraud and behavioral analysis through Amplify’s AI routes.
AWS Lambda: Executes serverless functions for fraud detection, cancellations, and dispute workflows.
AWS Step Functions: Orchestrates multi-step workflows for cancellations and disputes with retry logic and state tracking.
Amazon DynamoDB: Stores transaction data, workflow states, and AI analysis artifacts for persistent state management.

💻 Frontend & Frameworks

Next.js (App Router): Modern full-stack React framework for building the UI and API routes.
TypeScript: Enforces type safety and scalability across the entire codebase.
TailwindCSS + shadcn/ui + Radix: Creates a cohesive, accessible, and responsive dashboard interface.
Lucide Icons & React Hook Form: Simplify user interactions and data validation.

🤖 AI & Logic

Amplify AI Routes (Bedrock): Used for fraud analysis, behavioral risk scoring, and transaction classification. Custom Rule-Based Engine: Combines heuristic checks and AI results to produce explainable fraud alerts.

🔄 Orchestration & Data Layer

Amplify Workflows (Step Functions): Manages end-to-end flows for cancellations and disputes.
Amplify Functions (Lambdas): Modular business logic for fraud detection, action execution, and workflow coordination.

🧩 Supporting Libraries

Zod: Schema validation for API responses and data models.
AWS SDK v3: Provides typed interfaces for Bedrock, Step Functions, DynamoDB, SQS, and Secrets Manager.