Tufail Ahmed

I’m a Full Stack Engineer specializing in scalable, high-concurrency architectures, with a strong bias toward performance validation under real load. I architect and load-test real-time, edge-optimized systems to ensure they hold up during high-traffic spikes.

In AeroVoiceAI, I built a real-time WebRTC conversational agent using Node.js and TypeScript, delivering <1s response latency via a hybrid cloud/local dual-engine architecture. I optimized local inference to fit within a strict 4GB GPU VRAM limit, including deploying a 1.7B parameter LLM and downsampling 48kHz audio to 16kHz for hallucination-free STT, while maintaining continuous audio processing with zero event loop blocking using Worker Threads.

For FlashSale, I engineered a bifurcated serverless backend on Cloudflare Workers and load-tested it with Grafana k6 to simulate 100,000+ user traffic spikes. I implemented atomic inventory locking with Redis Lua scripts, achieving a p99 latency (<10ms) during load testing, and used patterns like a Circuit Breaker and Dead Letter Queue (DLQ) to prevent cascading failures and automatically refund locked inventory on write timeouts.

In BlinkChat, I delivered a hyper-local chat experience with low-latency WebSockets and Redis caching, serving the 50 most recent messages per room instantly from memory and reducing initial DB reads by 80%. I also containerized the backend via Docker for AWS EC2 deployment and implemented cursor-based pagination, secure HttpOnly cookie authentication, and a cron-less TTL service to purge ephemeral messages older than 2 hours.