What I've Learned
Over the past few months, I've grown tremendously as a full-stack developer by building and deploying real-world products and gaining valuable industry experience. Four milestones define this chapter — scaling an AI SaaS to 1800+ users, building an AI-powered GitHub code analyzer, crafting a real-time MERN chat app, and surviving a live PHP internship.
LastMinutePreparation — From Idea to 1800+ Users
An AI-powered exam preparation platform for CBSE students — built, launched, and scaled to real users.
LMP was built with a clear goal: help students prepare effectively in the last moments before exams using AI. This wasn't just a coding project — it was a real product with real users, real traffic, and real revenue. From handling performance issues to scaling backend systems and acquiring users organically, LMP taught me what it truly means to build in the real world.
CBSE students on platform
From Pro plan subscriptions
Production deployed & running
The Biggest Challenge — Slow AI Responses
Initially, the system processed AI requests synchronously — the user sends a request, the backend calls OpenAI, and the response returns after a long delay. This created poor UX and risk of users dropping off.
"Real-world systems fail differently than local projects."
The Breakthrough — Queue-Based Architecture
Redesigned the system using asynchronous processing: introduced Inngest as a queue system, moved AI processing to background jobs, and implemented frontend polling for results.
Performance — Redis Caching
- Redis (Upstash) for caching frequent responses
- Reduced OpenAI API calls significantly
- Faster response times & lower costs
Monetization & Access Control
- Razorpay payment gateway integration
- JWT-based auth + rate limiting
- Free → Pro tier with auto-downgrade after 30 days
AI Features Built
- Topper-style answer generation
- Topic summarization
- Smart question solving
- Chat with PDF
- Diagram explanation (GPT-4o mini / GPT-5.1 Pro)
What LMP Taught Me
- Designing scalable backends using queue systems
- Handling async workflows in production
- Redis caching for performance optimization
- Building and monetizing a real SaaS product
- Organic user acquisition through Instagram Reels
- Solving latency, API limits & user retention issues
RefynAI — Automating Code Review with AI
An AI-driven GitHub code analysis platform — built with a friend, scaled from scratch, live at refynai.org.
RefynAI is a production-grade developer tool my friend and I built together — handling everything from architecture to scaling ourselves. The idea was simple but powerful: connect to a GitHub repo, scan it for bugs and vulnerabilities, and automatically generate a pull request with the fixes. No manual debugging, no waiting for code review — just AI doing the heavy lifting end-to-end.
GitHub repositories analyzed
Bug-fix PRs auto-generated
Live at refynai.org
End-to-End Automation Pipeline
The entire flow — from repo scan to merged fix — runs automatically. We built an async pipeline using queue processing so even large repositories don't block the system.
GitHub Integration
- GitHub OAuth for secure repo access
- PR lifecycle management via GitHub API
- Auto-fix toggle for hands-free improvements
Global Payments
- Razorpay for Indian users (INR)
- PayPal for international users (USD)
- Pro tier with scan quota & advanced features
Analytics Dashboard
- Total scans, issues found, PR stats
- Per-repo scan history & issue breakdown
- Severity heatmap: Critical / High / Medium / Low
Tech Stack
The Hard Parts — What We Didn't Expect
Scanning large repositories is slow — we had to build an async processing pipeline from the ground up. GitHub API rate limits, PR conflict handling, and keeping the dashboard in sync with live scan state were the trickiest engineering challenges.
"Shipping a collab project to production teaches you how to build systems, not just features."
What RefynAI Taught Us
- Building complex GitHub API workflows & PR automation
- Async processing pipelines for large codebases
- Redis caching to handle high-frequency scan requests
- Dual payment integration (Razorpay + PayPal) for global reach
- Real-time analytics dashboards with live state sync
- End-to-end product ownership — from design to deployment
MERN Chat App — Real-Time Communication at Scale
A production-ready real-time messaging platform built with MongoDB, Express, React, Node.js, and Socket.io.
I wanted this to feel like a professional, scalable messaging app — not just a demo. The core challenge was implementing real-time communication without delays, alongside typing indicators, read receipts, file uploads, JWT auth, and a smooth responsive frontend. Debugging real-time Socket.io events, handling CORS, and maintaining database sync under load were the biggest growth moments.
Tech Stack
What I Learned
- Socket.io for bidirectional real-time communication
- Secure JWT auth with role-based access
- Efficient data modeling with MongoDB + Mongoose
- Media uploads via Multer & Cloudinary
- Deployment, CORS, and sync challenges on Render
Internship — From Zero PHP to Live Production
Joining a live production codebase, learning PHP in 3 days, and deploying a real eCommerce platform.
During my internship I was challenged to learn PHP in just 3 days and immediately work on a live React + PHP eCommerce project deployed via cPanel. I had to understand legacy backend logic, debug broken API endpoints, configure domains and SSL, and push production updates independently. Through constant trial and error, I improved performance, backend security, and deployment workflows — while keeping the live site running.
What I Learned
- Learning PHP from scratch in 3 days under pressure
- Integrating PHP backend with a React frontend
- Debugging large codebases and understanding legacy structure
- cPanel deployments, database config, and SSL management
My Journey Forward
These experiences changed how I think as a developer. I'm no longer just building features — I'm building systems that scale, perform, and generate real impact. LMP and RefynAI weren't just projects — they were products used by real users, systems that handled real challenges, and journeys that taught me more than any course ever could.