AdityaShibu
Most “AI engineers” wire up someone else's API and call it done. I write the model, run it on the device in front of you — computer vision in your browser, no server in the loop. Came up through C++ and systems. Going all-in on computer vision.
New Delhi, India · will relocate for the right chaos
What's not in the stack
Built like it's 1998, not a hackathon.
No cloud lock-in
If it needs a subscription to someone else's inference API to work, it doesn't ship. Models run where the user is.
No black-box APIs
No gluing together someone else's endpoint and calling it engineering. If I can't explain the pipeline, it's not done.
No untested claims
Numbers come from evals, not vibes. If a project says it works, there's a benchmark backing it up.
No bloated deps
Systems background means I reach for the smallest tool that solves it, not the trendiest framework of the month.
Choose your stack
Things I actually shipped
Six projects, six different stacks, zero tutorials followed to the letter. If it runs at the edge and doesn't need a server to feel alive, it's probably mine.
Why me
Not a checkbox hire.
Systems-first
C++, an emulator mid-build, a real-time anomaly-detection engine — before I ever touched ML. The foundation is real, not a bootcamp certificate.
Ships on-device
Not a wrapper around someone else's API. Vision runs in-browser via WebGPU and MediaPipe — nothing round-trips to a server to work.
Documented in the open
Every project's a public repo — real commits, real history, half-finished parts included. Nothing here is a private codebase you have to take my word for.
Oh, this?
Built for IBM ThinkFest 2026. Not bad for a systems guy moonlighting in ML.
Engineer No. 001
Not your average ML hire
C++ and systems first, then ML — in that order, on purpose. I build computer vision that runs on-device: in the browser, at the edge, with nothing phoning home to a server.
Started in C++ writing low-level things nobody sees — a Game Boy emulator that's still mid-build, a real-time anomaly-detection engine. Then ML got its hooks in me and didn't let go. Now I build models that actually run where you are: on-device, in the browser, no server standing between us. Computer vision is the deep end and I'm still swimming down.
PyTorch from scratch · CNNs · RAG + LangGraph · MLOps — nothing here is a checkbox
Python · PyTorch · C++ · TypeScript · FastAPI · Docker · Linux — no drag-and-drop, no low-code
Chipping away at a Game Boy emulator for fun and daily-driving Arch, because apparently things need to be hard on purpose.
Deep Learning (Goodfellow) · Designing Data-Intensive Apps · Grokking Deep Learning
New Delhi, India · remote · will relocate for the right kind of trouble
How we could work together
Pick a lane.
Full-time
One team, one problem, all the way through.
- ✦ML / CV engineering roles
- ✦Systems-adjacent teams
- ✦Relocation on the table
Contract
Scoped work — a model to ship, a pipeline to fix.
- ✦Fixed-scope engagements
- ✦Short research spikes
- ✦Remote, any timezone
Proof, not vibes
Verified, not vouched for.
Writing
Things I learned the hard way
Side Quests
OSS & tinkering
Short-form, shipped daily-ish
Longer writing lives above. The build-in-public updates, hot takes, and progress notes land on LinkedIn first.
Say something
Let's build something real
Hard problems, research rabbit holes, the right opportunity — I'm in. Drop a line, skip the small talk. I reply within 48 hours, not “within 2 business days.”