E-Learning Platform
Course platform with 50+ REST APIs, Razorpay payments, multi-layer caching and resumable multi-GB video uploads. 10K+ downloads on the Play Store.
Overview
A full-stack learning platform where students browse a catalog, buy courses, packages and events, and stream long-form video. Instructors upload content; admins manage catalog, pricing, promos and access.
The Problem
The platform had to serve read-heavy catalog traffic at low latency under high concurrency, take payments without ever double-charging or leaking access, and let instructors upload multi-GB videos over unreliable connections without restarting from zero.
My Role
Backend engineer — Code Brew Labs
- Architected the backend — 50+ RESTful APIs across authentication, course management, payments and admin, with role-based access control.
- Owned the entire payment flow on Razorpay: order creation, signature verification, automated refunds, promo codes, loyalty points and GST e-invoice generation for course, package and event purchases.
- Built a multi-layer cache (in-memory + Redis) with Pub/Sub invalidation across instances, cutting API latency ~60% under high-concurrency traffic.
- Designed asynchronous, event-driven pipelines on Bull with retry, backoff and dead-letter handling to hold up throughput and fault tolerance under load.
- Built a secure S3 multipart upload pipeline using presigned URLs and STS tokens for resumable multi-GB video uploads across unstable networks.
- Set up CI/CD with GitHub Actions and wrote Jest unit + integration tests to catch regressions before deploy.
- Built React frontend modules (Context API, memoization, Axios) with a reusable component architecture for course management and user dashboards.
Engineering Challenges
Latency under concurrency
Catalog and course-detail reads dominated traffic and spiked hard. A two-tier cache (process memory in front of Redis) with Pub/Sub invalidation keeps every instance consistent on writes and dropped API latency ~60%.
Payment correctness
Retries and duplicate callbacks can double-charge or grant access without payment. Every Razorpay callback is signature-verified and idempotent; refunds, promos and loyalty points all resolve against one order record.
Large, unreliable uploads
Instructors upload multi-GB videos on flaky networks. Multipart upload with short-lived presigned URLs and STS credentials lets a client resume from the last completed part instead of restarting.
Throughput and failure isolation
Email, invoicing and post-purchase work can't block the request path or lose events. Bull queues with retry, exponential backoff and a dead-letter queue absorb spikes and quarantine poison jobs.
Solution
A layered Node.js/TypeScript service: thin controllers, a service layer that owns business rules, repositories over MongoDB. Reads go through the two-tier cache; writes publish invalidation. Payments run through a single order record with idempotent callbacks. Background work is queued on Bull. Uploads are brokered as multipart sessions against S3. CI/CD runs tests on every push via GitHub Actions.
Key Features
- ›Course, package and event purchases
- ›Razorpay payments, refunds, promo codes, loyalty points
- ›GST e-invoice generation
- ›Role-based access (student / instructor / admin)
- ›Resumable multi-GB video uploads
- ›Instructor and admin dashboards
- ›CI/CD with automated tests
Architecture
- 01/React client (Context API, Axios)
- 02/REST API — Node.js + TypeScript, RBAC
- 03/Two-tier cache — in-memory + Redis (Pub/Sub invalidation)
- 04/Bull queues — email, invoicing, post-purchase jobs
- 05/MongoDB (primary data)
- 06/AWS S3 — multipart video storage (presigned URLs + STS)
- 07/External — Razorpay
Technology
Result
~60% reduction in API latency under high-concurrency traffic. Payment flow (orders, refunds, promos, GST e-invoicing) shipped end to end and running in production. The app has 10K+ downloads on the Play Store.