Follow the journey
Lab tour: where cameras learn to think
GPU racks, whiteboard sessions, and chai. Behind the scenes at Drik Vision AI in Guwahati.
View on Instagram
short Blurry plate to clear text in 15 seconds
DrikNetra multi-frame super-resolution — from surveillance-grade blur to readable license plate text.
This IIT Guwahati startup is building AI that understands Indian traffic
Drik Vision AI is building visual reasoning systems for traffic cameras, starting with the hardest roads on Earth.
Read article video DrikLabel: The Annotation Tool We Wished Existed
Full tutorial: import a video, auto-annotate with YOLO, correct a few frames, and watch active learning propagation do the rest.
Building an AI startup from Guwahati, not Bangalore
Everyone said we needed to be in Bangalore. We chose to stay close to the problem. Here's what that looks like.
View on LinkedIn video Watch AI Try to Understand Indian Traffic
Five levels of visual reasoning applied to a real Guwahati intersection. From raw detection to causal reasoning — in real time.
Why we built five levels of visual reasoning
Detection is Level 1. Most traffic AI stops there. We built four more levels — recognition, description, reasoning, prediction. Here's why each one matters.
View on LinkedInAnnotating 10,000 frames in 2 hours with DrikLabel
Active learning propagation in action. Correct one frame, watch corrections flow across the entire track — forwards and backwards.
View on Instagram