Portfolio • Email • X • LinkedIn

I spend most of my time learning about systems and figuring out why something that should work… doesn’t.

Currently building at Portkey AI.
Mostly working around APIs, infra, and the fun parts of networking that only show up in production.

I enjoy:

• tracing bugs across layers until they stop hiding
• making systems a little more predictable than they were yesterday
• removing unnecessary complexity (and occasionally adding the right kind)

• Eddie (private): a local voice assistant for macOS, inspired by JARVIS/FRIDAY. A Node.js agent with a native Swift layer for speech and audio, interruptible by voice, and integrated with Calendar, Reminders, Contacts and confirmed iMessage/WhatsApp sends. It answers in English and Hinglish, and reaches the iPhone through Siri Shortcuts over Tailscale.
• Braventex: a student-only community and collaboration platform with communities, a feed, study rooms and realtime notifications. Built as a modular monolith on Fastify + Mercurius (GraphQL), Prisma/PostgreSQL and WebSockets, with a Next.js 15 frontend.
• violence-detection: real-time violence detection on RTSP/webcam streams. A PyTorch MobileNetV2 classifier with Faster R-CNN and MediaPipe for people and pose, served over FastAPI WebSockets, with detections stored in MongoDB and clips uploaded to S3.
• devansh-portfolio: devanshdubey.com, built on Next.js with an MDX blog, RSS and email subscriptions, and interactive demos running C++ kernels compiled to WebAssembly.

Python · TypeScript · Go · Node.js · Fastify · Django
GraphQL · Kubernetes · Docker · Azure
PostgreSQL · MongoDB