SAMIR POKHREL · KATHMANDU
Backend engineering for agentic AI, MCP infrastructure, real-time media pipelines, and the operational layers that keep them stable once complexity shows up.
Systems, products, and tooling shipped for real use.
ALL PROJECTSNo-code agentic workflow platform. I lead the Gateway Service, Unified Auth, and the STDIO ↔ SSE proxy layer behind a fleet of MCP servers.
Event-management SaaS with a real-time media stack — RTMP, WebRTC, FFmpeg → HLS — plus a dynamic branded-app system that generates event-specific mobile apps from one dashboard.
Node + Express + TypeScript boilerplate generator. Cuts setup time ~60% with type-safe scaffolding, REST/GraphQL toggles, Prisma integration, and ready-to-run dev/build/prod scripts.
Notes on MCP, media pipelines, and systems under pressure.
READ THE BLOGA field manual for the parts of the Model Context Protocol the docs don't cover yet — gateways, proxies, transports, auth, and the small decisions that compound.
Webhook runners, polling engines, and why cron alone is not enough when you're orchestrating agents that need to survive failures.
Hard lessons from wiring distributed Node.js and Python services together with gRPC — schema discipline, error surfaces, and when to just use HTTP.
A real-time media pipeline I shipped at Eventsmo — what worked, what didn't, and what I'd change if I were starting over.
Queue design, worker lifecycle, and the failure modes you won't see until you're live — notes from running BullMQ under real load at Lasius and Eventsmo.
Systems at the boundary of products and infrastructure.
MORE ON ABOUTBuilding the backend that makes AI agents reliable — MCP server fleets, STDIO ↔ SSE proxies, durable orchestration with Temporal, and gateway routing that ties it together.
End-to-end media pipelines: RTMP ingest, WebRTC over WebSockets, FFmpeg → HLS/FLV transcoding, and S3 archiving — built to run live under real load.
Distributed architectures — gRPC between services, event-driven queues with BullMQ, unified auth layers, and REST / GraphQL APIs that scale without rewrites.
Containerised services on AWS, queue-driven background work, OpenTelemetry traces across MCP hops, and infra ops teams can reason about at 2 AM.
Backend engineering for teams that need systems to hold up. Send context — I reply directly.