Founder at Maniac
Dhruv Mangtani
B.S. Physics and Mathematics at Stanford\\ 2x founder\\ theoretical physics at Caltech\\ ML research at Berkeley\\ building Maniac day & night
Questions & Answers
What's the story behind your company?
I thought it was wild that people are repeatedly paying for trillion parameter, closed-source, generalist LLMs in order to solve narrowly defined tasks. Not only do businesses lose negotiation leverage over their LLM costs, but they also lose access to exciting research in compression, interpretability, RL, and other fields. The basic unit-economical inefficiency, the loss of IP to the AI layer, and the lack of technical control made me realize that a platform for transitioning from LLM vendors to proprietary models is inevitable. I want to build that platform.
Can you share a distinctive achievement or highlight from your career?
My first deep-tech startup in the Web3 space ultimately failed to find PMF, which taught me the value of unifying theoretical technical beauty with real-world market opportunities. It also taught me how much I loved simultaneously selling amazing products and doing cutting-edge research.
What sets you apart from others in your industry?
Others in the industry are point solutions for different possible techniques of optimizing the larger model-layer, such as routers, judges, fine-tuners, RL environments, context engineering, and prompt optimization. In practice, stitching together these tools requires a large upfront cost of engineering hours and continuous maintenance over time (as new features, bugs, models come up). These tools lose sight of the big picture: making it easy to build your own proprietary model layer with maximum algorithmic control. We provide connectors for all of these tools and orchestrate continuous CI/CD over your evals through our Agent Containers, allowing you to create your agents once and then always own a SoTA model layer. Think, Kubernetes for the AI Era.
