We first met Akshat Kannan when he was 17. We immediately knew that he was absolutely brilliant, a true outlier talent.

Akshat knew that he would not go to Stanford to study CS; instead, he wanted to work on Zeroset full time. His vision, his raw ambition, was so compelling and so clear that we had to get involved.

Akshat's co-founder, Will Zhang, is equally brilliant. Before teaming up with Akshat, he dropped out of UT Austin, built production RAG pipelines at Amazon, and founded the retrieval startup Fynopsis.

Together, Akshat and Will invented a better, more traceable memory for AI.

Their first product, Nebula, pulls together a company's tickets, messages, approvals, handoffs, tool calls, and decision records into a live model of the business: what's happened, what's changed, what's still open, and how the pieces connect.

Nebula structures information the way people do. Instead of reconstructing workflow state from raw data and chat logs on every run, agents can just ask Nebula for it.

The benchmarks are strong: on LongMemEval, Nebula beat Mem0's retrieval accuracy by 20%, with sub-50ms p50 retrieval at lower token cost.

Retrieval is only the starting point. Zeroset plans to train customer-specific models on these structured traces: "enterprise world models" that learn how a business evolves and which actions move it forward. The setup mirrors robotics: a learned policy picks the next action, and lower-level controllers execute it. Here, frontier models do the executing while the world model decides what comes next.

Today, Zeroset announces its $5.2M inception round, led by 2048 Ventures and Gradient. We could not be more stoked and proud to back Akshat and Will and support this exceptional company along with our friends at Gradient.