Karthik Ravi

How I got here

I was an ML engineer at Facebook. Left, started companies, and had to figure out sales the hard way.

At Cue Labs, we started with tools to help companies sell. Then we put AI agents to work ourselves, and our OpenAI bill went past $40,000 a month. That sent me down the stack: into inference, models, and the hardware underneath them.

I started with a question about cost. I kept coming back to a question about control: what would it take for a team to run powerful AI on its own terms?

The work I want AI to reach

Think about an engineer working on a protected defense program, or a biotech team building software around proprietary research. The code, context, and knowledge that would make an agent useful are also the things they need to protect.

A model is only useful if you can bring it into the work. For these teams, that means running inside their environment, connecting to their tools, and working within their operating rules.

I don't think teams should have to choose between protecting their knowledge and putting it to work with AI. The people building critical systems and pursuing new discoveries should be able to use these tools where their work already lives.

What we're building

CueCloud

CueCloud is how we're working toward that. We're bringing the compute and the agent workspace together, starting with defense engineering and biotech R&D. Models running on your hardware. Agents working with your code and tools. Your team in control of the environment.

We work alongside your engineers to make it useful: size the hardware, configure the models, connect your tools, and validate the deployment on real tasks. My goal is for a team to spend its time on the engineering or discovery it came to do, with AI running inside the boundary it needs to keep.

Thoughts

I write The AI Company Builder Memo. How to build and sell AI companies.