Unicorn AI Summit.
Engineering talks from the teams behind India's leading unicorns, alongside live AI startup booths.
21 Sep 2026 · Bengaluru, Karnataka
The evening opened with Sidhant Panda of Swiggy on forking a business: how to model a company from first principles, turn its metrics into a tree an agent can read, and make that agent the first responder when a metric misbehaves.
youtubeForking the Business
Sidhant Panda · Swiggy
Venkatesan Vaidhyanathan, Chief Architect at Slice, followed with how a fintech turned regulated bank puts AI to work: a control layer that keeps company data and credentials from leaving, and AI code review fed with its own context.
youtubeTrust Is Not a Feature Flag
Venkatesan Vaidhyanathan · Slice
Devinder Gupta of Flipkart then asked what agents actually need to know, separating the code plane from the knowledge plane.
video coming soonCode Plane, Knowledge Plane: What Agents Actually Need to Know
Devinder Gupta · Flipkart
Adarsha Das of Meesho took on the cost of running LLMs behind production agents, with a model that reasons in state space instead of token by token, splitting reasoning into independent steps that need no words.
youtubeThinking Without Words: Reasoning in State Space, Not Token Space
Adarsha Das · Meesho
Tirumarai Selvan of PromptQL asked what stops teams from using AI for all of their work, and made the case for multiplayer AI: people and bots working together, so every lesson is kept and the organisation's context gets smarter each day.
youtubeMultiplayer With Bots Is the Future of Work
Tirumarai Selvan · PromptQL
Anshu Avinash drew on three years of production agents at DevRev to show what a harness needs beyond the model loop: tools, permissions, crash recovery, streaming and managing context over long runs.
youtubeAgent Harness and Evals
Anshu Avinash · DevRev
The booths (48)
Partners
- PromptQL
- Grayscale Ventures