Fintech PWA: Bank Fee Disputes Cut to Under 1 Second

The core backend used Django REST Framework for bank accounts using Plaid, virtual cards using the Lithic API, BNPL payments tracking, and transaction history. A different Django service dealt with authentication using CognitoHQ and phone login using Twilio. A Django admin was used for managing KYC, moderating users and ML model fine-tuning. The frontend was a React PWA rebuilt around RTK Query, while the landing page used Next.js and Framer Motion.

The team replaced the hooks in the frontend API layer with RTK Query to standardize caching, invalidation and loading state for all components. Service Workers were implemented to cache the transaction history and subscription data when offline. Deep memoization was done across the React components and Redux selectors to remove redundant re-renders. Firebase FCM push notifications as well as scroll-based animation using Framer Motion on the landing page were implemented by the team. The ML fine-tuning admin panel and additional endpoints for the Django REST API were developed.

The platform was based on AWS and used SNS, SQS, Lambda, S3, EKS, EC2 and ECR. EKS ran the containers on EC2 compute, S3 handled storage, ECR stored the container images, Lambda ran serverless functions and SNS handled pub/sub messaging. The message queue was managed by AWS SQS, and the applications were packaged into containers using Docker. CI/CD ran through GitHub Actions and Jenkins, which automated builds and deployments. Notifications were delivered through Firebase FCM.

KYC verification was done using CognitoHQ via the auth service, while authentication via phone was done using Twilio. Feature flags were used to limit the exposure of new features until rollout. Connectivity with banks was provided via Plaid, which is a regulated financial data integration. Virtual cards were issued using the Lithic API.