Varad Parate

a production-grade monitoring platform that performs automated API health checks at configurable intervals, tracks 24-hour rolling uptime and response latency, and alerts users only when status changes (UP ↔ DOWN).

Key engineering decisions:

• Built a background monitoring worker with node-cron, limiting concurrent requests to 10 for stable CPU/memory on AWS EC2.
• Automated alerts via Nodemailer — fires only on status transitions, not every check.
• MongoDB TTL indexes auto-delete logs older than 30 days.
• 4-job GitHub Actions CI/CD pipeline running 49 automated tests (unit + integration) before every deployment.

Deployed React frontend on Vercel, Express backend on AWS EC2 with Docker Compose and Nginx.

a CLI-based HTTP caching proxy that intercepts GET requests and serves them from an in-memory cache, achieving O(1) lookups with JavaScript Maps.

• Reduced API response times by over 99% (from ~440ms to near 0ms) for cached requests.
• Custom LRU eviction and TTL expiration logic to manage memory efficiently.
• Smart proxy routing safely bypasses the cache for data-modifying requests (POST/PUT).
• Graceful shutdown protocols prevent active connection drops during container scaling.

Dockerized and deployed to Render for live production traffic.

a server-authoritative real-time quiz engine where students battle in timed 1v1 matches across OS, DBMS, CN, and OOPs — designed for competitive placement preparation.

• All match state lives in-memory using JavaScript Maps, reducing database writes from ~20+ per match to exactly 1 batch write at match end.
• Elo-based matchmaking with progressive queue expansion (±50 → ±350 Elo range) and an intelligent bot engine that injects skill-calibrated opponents within 15 seconds.
• Server-authoritative scoring — correct answer indices are stripped from payloads before reaching the client, making DevTools cheating impossible.
• Custom Socket.io reconnection with a 5-second grace period: disconnected players rejoin seamlessly, or auto-forfeit if they don't return.

Ships as a single Docker container — Express serves both the API and the compiled React frontend, eliminating CORS and multi-service overhead.

a next-generation learning platform that uses Google Gemini AI to generate adaptive educational content on demand — no pre-collected datasets, no static curricula.

• Learners define their topic, difficulty level, and academic goal; the AI generates a full syllabus, lesson notes, quizzes, flashcards, and Q&A in real time.
• Built on a serverless stack: Next.js frontend, Neon PostgreSQL for persistent storage, Inngest for background AI orchestration, and Clerk for auth and billing.
• Content is validated for JSON/HTML structure and relevance — no hallucinated formatting, every quiz has verifiable answers.
• Mobile-responsive, browser-based, and low-bandwidth friendly — no downloads required.