Twitxy Saved

Sepand Dyanatkar

Verified account This account is verified. Learn more

@SepandD

@OnDeckAI (YC S25)
Grounded visual reasoning at scale. Co-Founder and CTO.

Master’s in CS @Cambridge_CL
Canada linkedin.com/in/sepandd Joined November 2011
319 Following 240 Followers 153 Posts

Latest posts 5

Pinned post

Sepand Dyanatkar Verified account @SepandD
Really excited to tackle this head on.

We've taken it upon ourselves to publish clean benchmark results for the world to use, and ultimately build out new standards for vision.

We’ve been busy building & deploying great real-world vision systems, and assumed someone else would take this on. But no one did, so here we go!
OnDeck AI Verified account @OndeckAI Sep 23
We are excited to have released the first set of evals on VisionIndex. Below, we shed some light on why we’re building this and why we’re opening it up to the world. 🧵 1/n
View quoted post on X
4 replies 2 reposts 12 likes 628 views View on X
Sepand Dyanatkar Verified account @SepandD
Do video benchmarks actually test video understanding? Excited to announce our NeurIPS 2026 workshop paper investigating problems with the state of video evals, and ways to address them.

One of the many problems is highlighted in the image below (option-conditioned blind solvability)
Sepand Dyanatkar Verified account @SepandD
Human feedback helps capture the minutiae of hyper specific vision tasks. We added an arena to VisionIndex to capture and show them to users.

Vision Arena is live! Try it out.

More models coming soon.
OnDeck AI Verified account @OndeckAI Nov 10
Vision benchmarks are blind to subtle preferences.

Now you can bring your own video and imagery to frontier models, and see how they perform on your tasks. The Arena is live today on VisionIndex.
View quoted post on X
Sepand Dyanatkar Verified account @SepandD
Evaluating vision capabilities is heavily multidimensional. Our users need to often know what the cost of every additional percentage point of accuracy is.

Now, for public benchmarks everyone can see this on VisionIndex.

We've also added Gemini 4 Argon, which tops the leaderboard. Let us know if there are any other models you want to see.
OnDeck AI Verified account @OndeckAI Nov 9
Measuring vision performance needs more than just accuracy.
We've released cost and latency axes on VisionIndex.

Now you can explore per-task performance and efficiency across vision models.
View quoted post on X
Sepand Dyanatkar Verified account @SepandD
Control theory is a great analogy.

Also: Eerily similar to agentic engineering findings, where the rate of feedback, essentially what guarantees stability, is the speed limit.

It also reminds me of policy optimization in RL

Always fun when a new type of task+problem emerges and we’re finding so many connections to things that we’ve already worked on.
Peyman Milanfar Verified account @docmilanfar Sep 12
Why you can’t do RSI as fast as you like - x.com/i/article/2098…
View quoted post on X