Fireside chat at the NVIDIA & AWS Startup Mixer

About the event

The NVIDIA & AWS Startup Mixer was an invite-only afternoon for startups working on physical AI. NVIDIA presented it with AWS Startups. It ran from 3:00 to 6:30 pm IST at the NVIDIA office in Bengaluru.

NVIDIA's invitation said: “Hear from Stellarview in a fireside chat on real-world Physical AI deployment challenges.”

The fireside chat

Niranjan Patil, co-founder and CEO, and Mayur Kulkarni, co-founder and CTO, took questions in a fireside chat. Arundhati Banerjee, Startup Ecosystem Lead for South Asia at NVIDIA, moderated it. The chat covered what happens when AI is deployed in the field and what the two founders learned on the way.

Before the chat, NVIDIA and AWS presented their startup programmes and their technical view of the physical AI stack. The afternoon ended with time for startups, investors and partners to meet.

What we talked about

Our starting point: a model that works in the lab is only the beginning. Our post for the event listed four problems that real deployment brings:

• changing light and weather
• unpredictable scenes
• edge hardware limits
• the need to run 24/7 without missing a beat

On Indian roads each of these is daily work. A camera on a pole has a fixed compute and power budget, it sees mixed traffic in every light, and nobody is on site to restart it.

How our products answer these

The Recon Global Shutter Camera captures the full frame at once, so fast vehicles do not smear, and uses Pulse IR for night capture. Its models run on an NVIDIA Jetson Orin NX inside the camera, so a weak network link does not stop detection. EdgeTrack monitors the uptime of devices in the field.

For more on how we test models before they reach a road, see our research page.