Prajyot PORJE

I’m a full-stack engineer who independently builds and ships production AI systems—multi-tenant MCP infrastructure, self-healing multi-model LLM pipelines, and browser-native execution environments adopted by users internationally.

In my recent work, I built DevFlow, a fault-tolerant AI code generation and execution platform with a 5-model OpenRouter fallback chain and automatic API key rotation, achieving 99.2% generation uptime for 25 users and 36 production workspaces. I also reduced browser-native preview load time from ~5 minutes to under 60 seconds, and engineered a Babel-AST validation layer with a Gemini-powered 2-attempt self-repair loop to drop preview crash rate from ~12% to under 0.5%.

I’ve also delivered ContextGraph, a production multi-tenant MCP server (JSON-RPC 2.0) that safely scopes developer context with SHA-256-hashed API keys and zero-trust PostgreSQL isolation at the API layer to prevent cross-tenant leakage. Alongside building features, I enjoy debugging and improving developer experience—like fixing WebGL pointer-event desync issues in a 3D context visualization—and I’ve contributed to open source by merging a JupyterLab PR.