Arena Log

This was through EF def/acc cohort. Worked with an Oxford PhD to turn causal AI into a company. We explored several use cases and landed on root cause analysis in manufacturing. After having built a V0 and 80+ interviews with potential customers and leading scientists we acknowledged an existing market unreadiness and tech immaturity. There's lots to do still to help manufacturers leverage basic data on one hand, and lots of open questions to answer in Causal AI research on the other. We decided not to fundraise as we could not see a promising direction towards scalable PM fit.

I worked with a PhD group and Professor from University of Trieste to develop a software platform for the NISQ (Noisy Intermediate-Scale Quantum) computation era. Our focus was on building tools for noise modeling and error mitigation, which we saw as critical needs in the quantum computing space. After months of iteration on narrative and initial experiments, we shelved the project due to the very early stage of R&D. Quantum is still very very early.

Autonomous robots for room service delivery in hotels and food service delivery in restaurants. We were pretty early with this. Completely boot/grant-strapped by design. Built prototype, obtained LOIs, presented at CES in Vegas, and other things that made me learn a lot. I left to chase my youth dreams of a world changing VC-backed company. TactileRobots is still there and Riccardo is doing a great job at making progress.

Full-service fast casual automated restaurant franchise. The concept was to introduce consumer-facing automation In the food industry via starting a comfort food franchising chain. We positioned targeting vegan or vegetarian friendly consumers selling plant-based burgers. An autonomous robot would then serve the customer at their table. This idea attracted quite some interest when pitching to investors. I then decided not to pursue this idea because it basically meant we had to start 2 businesses in 1 (fast food, and robotics) and I focused on the robot first with Tactile Robots.

Smart Wheel Impact Load Detector. Predictive maintenance algorithmic solution for railway wheels. Leveraging existing hardware infrastructure, SWILD could measure the type and severity of railway wheels defects earlier and more accurately thean existing systems, leading to better predictive maintenance and better preservation of the infrastructure. Project started with a TU Delft PhD. A POC with the Dutch railways fell through after months and I focused on graduation.

We wanted to use tDCS (Transcranial Direct Current Stimulation) to treat mild depression + give patients an app to connect with doctors and monitor the therapy. We built a physical design prototype for the device, a prototype for the app, a deck, and we talked with 50+ medical doctors, patients, and clinical researchers. We decided to abandon the project when it was becoming clear from our interviews and studies that the scientific evidence for our specific use case (mild depression) was dubious.

Automated grocery reception for perishable goods delivery. Together with 3 friends, we designed a concept for system of modular refrigerated boxes (similar to an impost collection point), that a grocery delivery service could use to deliver groceries without the need for the customer to be at home. We ultimate the conceptual mechanical design of the system of boxes, put together a deck and started pitching. Project shelved when 2/3 of the team departed for study abroad. Also, we had no clue about what we were doing.

I have been a member of the first def/acc cohort. It was inspiring, some of the key takeaways I got are about keeping ambitions high, and the fact that you can probably accomplish much more than you think in 1 week.

During my first months in SF, I helped the Xoogler community setting up and run their first startup program in Palo Alto. This was very fun, learned a ton. I got to select startups, match senior mentors with founders, and sit in founders-VC pre-seed meetings. It was a truly great experience where -among other things- I appreciated the relevance of building a strong network.

I participated in the YES!Delft Validation Lab twice, first with SWILD and later with TactileRobots. Very useful program to get access to resources, great mentors and solid industry contacts for prospects.

Searching for real estate properties sucks! I felt it would be much cooler to use natural language queries and ask for "A bright penthouse with ocean views, spacious living room and a kitchen with marble". So I built a minimal prototype. It was an excuse to dig deeper into multimodal semantic search (it is a hard thing to do well!)

Sit and watched Mr. Karpathy videos on LLMs and felt it was useful for me to go through his tutorial slowly and step by step rebuilding it. Turns out it might be useful to others: sharing is caring.