Manish joshi

I’m a Software Developer focused on building secure, production-grade mobile and backend systems with measurable performance wins. Most recently, at Esferasoft Solutions, I architected “Meng AI,” an AI fitness SaaS on Flutter + Node.js + Gemini API, achieving sub-300ms average response latency under concurrent load.

I also build revenue-critical workflows end-to-end—at Esferasoft, I implemented a secure iOS & Android In-App Purchase flow with server-side receipt validation, reducing payment failure rate to under 1% across all subscription tiers. Before that, as a self-employed developer, I trained and deployed a YOLOv8 model for real-time object detection (optimized to 11MB TFLite for on-device inference) and delivered 8+ production applications for international clients.

On projects, I’ve taken full ownership of scalable AI infrastructure: SnapQuote AI (multi-tenant Node.js + Supabase with row-level security, JWT auth, and tiered access) and a “Semantic LLM Cache Proxy” that eliminated 100% embedding API costs and ~200ms inference latency using Xenova/all-MiniLM-L6-v2 via Transformers.js. I’m energized by distributed systems and reliability work—like Redis-based locking, ANN search with pgvector, and microservices ride-matching with Uber H3—validated through k6 load testing.