About · Jay Dodia

Hi, I'm Jay.

I'm a Computer Vision Engineer who genuinely lights up when research turns into something people can actually use.

My work lives at the intersection of ML research and production engineering. Day to day, that means designing real-time detection pipelines, fine-tuning state-of-the-art models like YOLOv26, SAM2, and MMAction2, and getting them to behave on messy, real-world camera feeds with imperfect angles and imperfect lighting.

I started out in cloud and DevOps — Terraform, Kubernetes, CI/CD, all the infrastructure-as-code plumbing — before falling hard for computer vision. That dual background is the thing I'm most grateful for. It means I can ship a model, but I can also ship the system around it.

Right now I'm leading a stealth Proof of Concept at Walmart for AI-driven theft prevention across Chile, Mexico, and Canada — owning strategy, architecture, and implementation. Before that I was building production person, cart, and action recognition pipelines at Sam's Club.

Off the clock, I'm usually hiking somewhere new, chasing a shuttle across a badminton court, designing something in a graphics editor, or recording my podcast.

Quick facts

The short version

BASED IN

Dallas, Texas Originally from India

CURRENTLY

Walmart Computer Vision Engineer

EDUCATION

Drexel University MS Computer Engineering, 3.5 GPA

LANGUAGES

English, Hindi, Gujarati + fluent in Python

CERTIFICATIONS

AWS CP · AZ-900 Cloud fundamentals

OPEN TO

CV / ML roles Full-time · Remote · Hybrid

What drives me

How I think about the work

The principles I try to operate by — in engineering, in teams, and everywhere in between.

Research meets reality

A paper is a paper until it runs on a live feed at 2 AM without crashing. I build for the real world — low-quality cameras, compression artifacts, variable lighting, and all.

End-to-end ownership

I don't stop at the model. Data pipelines, MLOps, deployment, monitoring, human-in-the-loop feedback — I want to be useful across every layer of the stack.

Partner, don't silo

The best CV work I've done has been cross-functional: Detection, MTMC, MLOps, Hardware, Action Recognition. Teams that share context ship better products.

Bias for clarity

Good documentation, clean code, simple diagrams. If a teammate can't pick up where I left off, I haven't finished the job.

Make it fast, then make it right

I ship ugly v0s early so the team can see, react, and steer. Perfection in isolation is a trap — feedback loops beat solo genius every time.

Curiosity compounds

The fastest engineers I know don't just work hard — they're relentlessly curious. I try to learn something that makes me a little dangerous every week.