VISHESH TRIPATHI

• Built a real-time multi-camera warehouse-vision platform in a pnpm/Turborepo monorepo (mobile, server, viewer, shared TypeScript packages) – React Native (react-native-webrtc) smart-glasses publishers and RTSP CCTV feeds (MediaMTX WHEP gateway) streaming into a Next.js monitoring dashboard over a Node.js/Socket.IO mesh signaling server with multi-viewer fan-out; deployed via EAS Android builds, AWS EC2 (Node signaling/vision server), and Vercel (dashboard).

• Engineered a server-side vision pipeline: cloud YOLO detection, zxing QR/barcode decoding, IoU object tracking with ghost revival, and a motion-event engine driving an event-gated VLM stage (Gemini) with a DBNet ONNX text gate – extracting license plates/labels only on tracked state changes, cutting VLM API calls ∼10× vs per-frame prompting.

• Shipped live 3D mapping on a cloud GPU – phone video to a RunPod RTX 4090 running a 1.2B-parameter feed-forward 3D reconstruction model behind a FastAPI microservice; the pod joins the Web