Video tools with an AI edit engine behind them. Perception runs at 500+ FPS on under 6GB of VRAM.
GPU, TensorRT, YOLO, FFmpegcaptos.pro
Hey, I'm Yash Jangid.
I like making software that's useful, fun to use, and worth sharing. I build tools for myself and others, and contribute to projects I find interesting. I've played around with everything from game development and 3D graphics to the web, backends, and AI. Over the last few years I've worked strictly on tools, backend systems, DevOps, and machine learning & AI. My projects have reached 300k+ people so far.
Right now I'm a computer science student. Previously at PhotoGPT, currently building captos.pro and Kensaku.
Products
Kensaku257
A local-first file launcher: search 100k file names in a millisecond, read inside PDFs, docs, code, and screenshots (OCR), and find by meaning. 250+ stars. Shipped via Homebrew, Scoop, and winget; listed in awesome-rust. Contributed as a corpus on XERJ Hub for AI agents.
Send a file device to device, with no upload, no timer, and no per-byte cost. 6.3K people used it last month.
WebRTC, P2P, Next.js, Goairdelivery.sitesource
Cobbicshut down
A realtime social app that held chat under 100ms with 10K+ users. I killed it, and here is why.
Realtime, pub/sub, Terraform, Typesensepost-mortem
Experience
PhotoGPTJan 2026 – Sep 2026Software Engineer Intern · Remote
Owned backend systems for an AI product-visual platform, covering, webhooks, rate limiting, caching, and production infrastructure.
Writing
EverythingCobbic had pages acting as places, a clean editor, real-time chat, search, and a feed that ranked conversations rather than followers. I built it in two months, deployed it on AWS, and let people in. Then I killed it.
A playful experiment where I trained a neural network autoencoder to learn a compact image representation, achieving up to 20× compression while preserving visual meaning.
AirDelivery lets you send a file from one device to another. No account, no server upload, no timer. Open the site on both sides, share a flight code, and the file travels straight over a WebRTC data channel.
I built a 3D SPH fluid solver in Unity using C#, Burst, Jobs, a spatial hash grid and instanced rendering. Drop a column of particles in a tank and watch it turn into a puddle.
More Projects
A fully on-device voice-to-voice pipeline — Faster-Whisper in, local TTS out.
A 3D SPH fluid solver in Unity, fast enough to watch a column of particles settle into a puddle.
A C++ particle sandbox built to see how many bodies the physics holds before it gives out.
A multi-client TCP chat room on raw sockets and threads, admin channel included.
A 3D renderer written from scratch, with no graphics API.
Unbeatable tic-tac-toe — perfect play means the best you can do is a draw.
MS lesion classification and segmentation from MRI scans, shipped as an offline Docker app.
Contact
Open to ambitious teams and ideas. Reach out via Email or X.