Home

Hi, I'm Rajat 👋

I'm a Software AI Embedded Engineer / Solutions Architect at an NVIDIA Partner Company — bringing modern computer-vision and vision-language models to run on tiny hardware. Jetson boards, microcontrollers, single-board computers: if it has a chip and a camera, I want the model running on-device.
Currently building: LocateAnything-3B on Jetson Orin Nano — NVIDIA's open-vocabulary detection VLM, running live on an 8 GB Orin Nano at ~25 fps with an optical-flow tracker carrying detections between the ~0.3–0.6 fps inferences. Full deep dive (architecture, throughput math, and every fix with its root cause) in the project post.
What you'll find here: projects I've shipped, deep dives on getting big models onto small boards, and short fix-notes for the sharp edges I hit along the way.
 
notion image

🚀 Featured Projects

Deep dives on getting production AI onto edge hardware.

📝 Recent Articles

Longer-form writing on edge AI, model optimization, and embedded systems.

🛠️ Edge AI Boards

Field notes on the boards I work with — specs, quirks, and what runs well.

Let's connect

▶️ YouTube: @atomicrajat