Ajay Gopi

Portfolio's a work in progress — just like everything in life, one commit at a time!

prof_pic.jpg

Ajay is a Visual AI researcher specializing in interpretability and performance optimization, with over five years of engineering experience building and scaling AI products at B2B startups. He is currently pursuing graduate research in Artificial Intelligence at the Rochester Institute of Technology (RIT), where his work focuses on practical strategies for deploying trustworthy and efficient AI systems in the real world.

Much of Ajay’s engineering journey has revolved around scaling edge-based computer vision systems under tight hardware constraints (and even tighter budgets). In one of his largest deployments, he orchestrated 3–4 GPU-optimized models per site across up to 225 locations, each handling 4 to 25 live camera feeds. At peak, this meant nearly 1,000 video streams running in parallel at 5 frames per second. That’s a lot of pixels… It sounds impressive—until you realize it’s basically babysitting 1,000 digital toddlers that occasionally throw tantrums when the network hiccups. Distributed chaos, in other words. It’s all about perspective.

Through these deployments, he’s come to appreciate that real-world bottlenecks don’t just stem from model size or GPU memory—but from overlooked constraints like NVIDIA’s video decoder throughput. Lesson learned: you can often load a seemingly massive model onto an edge device, but you can’t out-optimize the video decoder bandwidth. It’s like dropping a Ferrari engine into a golf cart—technically possible, but the wheels will still be your limiting factor.

Ajay’s current research draws heavily from these production war stories, with a focus on interpretable vision models, edge efficiency, and scalable deployment pipelines. He firmly believes the best AI systems are the ones that work reliably at 3 AM when no one’s around to restart them—a philosophy shaped by too many emergency calls asking, “Why did the cameras stop working?”

If you’re curious about trustworthy vision models, efficient edge ML, or just want to swap GPU war stories—feel free to reach out!


news

May 22, 2025 First submission to NeurIPS, details will be posted soon! (Oops, rejection)
Mar 20, 2025 Acceptance to AWARE-AI Research Traineeship (NRT)

latest posts