fluffbotics

fluffbotics was probably the first time everything I had learned over the previous decade came together in one place. I joined people I had known since school, assembled a small team across product, engineering, AI, growth, and eventually hardware, and got to work on problems that were difficult enough to require almost every skill I had accumulated. We started with an ambitious idea around LLMs and human consciousness. I built the early products and MVPs around that direction, and that work got the company a $3 million raise through a Google-backed fund.

Those ideas eventually became conversational AI products used by around 25 million people, then creator AIs, and later physical AI products. We built much of the underlying AI stack ourselves, learned how to run it at consumer scale, moved into electronics and manufacturing, sourced components in China, built on-device AI, brought inference costs down from roughly $45 to $5 per user per month, filed patents, exhibited at the Nuremberg and New York Toy Fairs, and secured a deal worth roughly $1 million. I was leading product, engineering, and growth while still being close enough to the work to sit down and solder prototypes myself.

What I remember most fondly, though, is the team. The office felt like a group of friends building something together. We travelled together several times, and those trips are some of my favourite memories from the company. With a large group, people would constantly split up, change company, explore in their own way, and come back together. I earned the respect of people I respected myself, and I learned a lot from the founders about the difference between thinking about a product and thinking about the company around it. I was often trying to make the product better. They were solving the larger strategic problem of where the company itself needed to go.

The company eventually ended for reasons unrelated to the work or its traction. I am at peace with that. Almost everything we set out to build, we built well, and the goals that mattered to us were achieved. More than anything else in my career, fluffbotics gave me confidence that the things I had spent ten years learning actually worked when I put them together. I am proud of the company we built, proud of the people I built it with, and happy that I got to experience it.

What we built

AI companions reached around 25 million users. The work included an in-house AI stack, consumer-scale infrastructure and on-device interactive toys.

Delivered work and results

• Built the early product and MVPs that got the company a $3M raise through a Google-backed fund, then turned that work into AI companions for Josh that reached 25M users with ~30 minutes of daily use and 45% D30 retention.
• Built the AI stack in-house across datasets, fine-tuning, and RAG, and scaled the infrastructure to sustain ~178K POST requests per minute.
• Evolved the product into creator AIs, then pivoted into conversational AI toys with on-device interactive AI.
• Reduced inference costs from roughly $45 to $5 per user per month and filed multiple patents around the underlying technology.
• Exhibited at the Nuremberg and New York Toy Fairs and secured a deal worth roughly $1M.