Read the longer story
Early entrepreneurship
During the COVID lockdowns and my final year of high school, I co-founded Faktor30[14]. Our first real client was a bank in Hamburg that needed a system none of us had built before. Our six-person team went on to build software for the bank, some other local companies, and we ended up creating a food-delivery company called Best Bite.
The experience also exposed a knowledge gap. I could make software work, but I did not yet understand why the underlying systems worked. I went to university to close it.
Theory and systems at scale
I studied computer science at TU Munich[12] to build the theoretical foundation in math and computer systems that coding alone had not provided. I then joined Snowflake's Database Connectors team[10] and later Microsoft[11] to learn how data and cloud systems behave at enterprise scale. The work showed how quickly assumptions fail when a system grows from a few users to many (larger scale) organizations.
I learned how to combine academic theory with practice at scale, but kept noticing that the people who invented foundational systems approached problems differently. I wanted to learn not only how to extend a system, but how to create a new one. That motivation led me toward research.
Academic research
I pursued my master's in computer science and spent a year studying and researching at UC Berkeley[13]. At the Sky Computing Lab[1], I worked closely with Ion Stoica[2], Joseph E. Gonzalez[3], and Matei Zaharia[4]. Through Matei, I also worked with Deepti Raghavan[5] and Phil Levis[6] at Stanford's Future Data Systems Group[15].
I went there to learn how to deconstruct systems, challenge assumptions across system layers, choose research problems that matter, and keep exploring when the path forward is not obvious. I'm very grateful for their efforts and mentorship.
One project focused on batch data analytics with LLMs, which was often framed as a model or hardware problem. The inefficiency was in how data flowed through the system. Reordering rows and fields in database tables increased KV cache reuse at the GPU layer without changing the model or hardware[7].
That work broadened into vCache[8], a semantic cache with formal error-rate guarantees, vAttention[9] for verified sparse attention, and SkyLight[16] for efficient inference. It also included compound AI orchestration through ALTO[17] and research on overthinking in agentic systems[18]. One of the learnings was to trace a problem across the entire system instead of optimizing one component in isolation.
I wanted to apply that way of research where the system did not exist yet, and beyond large language models (LLMs). That led me toward computational fluid dynamics.
Foundation models for computational fluid dynamics
I joined UniversalAGI's founding team[0] and helped establish our ML research effort. We research foundation models for computational fluid dynamics, where state-of-the-art simulations can require specialist knowledge and take hours to weeks. Once trained, our models produce high-fidelity computational fluid dynamic simulations with sub-minute inference latency.
These models are not LLMs. A single training sample can be tens of gigabytes, whereas LLM training samples are typically measured in megabytes, so many standard architectural and training assumptions break. I contributed to the training core and multi-node pre-training and fine-tuning cross-cloud infrastructure, and to the tokenization design of our LIFT architecture. I scoped and trained models such as SUV-PT, Rocket-PT, and Ship-PT on petabyte-scale datasets. The resulting models outperformed baselines including AB-UPT, Transolver, and GeoTransolver.
The research has also taken me into 3D geometry embeddings and morphing, image-to-3D and multi-fidelity simulations for lower-cost data generation, adaptive sampling as a multi-armed bandit problem, and data splitting and normalization across physical domains and datasets. The research sits at the intersection of model architecture design, (efficient) data generation, and distributed systems for training and serving.