Zurich, Zurich, Switzerland
Get in touch if I can be of help! Currently, I work on the intersection of Artificial Intelligence and Crypto. Previously, I co-founded Skilllab to gamify career transitions, similar to skill-trees in games. Initially motivated by the 2015 European Migration Crisis, Skilllab engages with public and private employment services and is used by millions of people, with the goal to navigate people through career transitions. I've worked on LLMs since 2018, fine-tuning and pre-training BERT and multi-modal language models from scratch, researching how they can be deployed at scale during an internship at Baidu research, and publishing one of the first papers on whether LLMs capture the notion of “meaning”. We were fortunate enough to earn the ~$1M Google AI Impact grant for my work at Skilllab, where we built the early recommender engine using matrix factorization techniques, including semantic embeddings, as well as transformers for collaborative-filtering. Before that, I was working on the ETH Hyperloop project, building out the first team and convincing stakeholders to provide us with resources (such as an airplane hangar). The project was integrated into the ETH B.Sc. programme after achieving 3rd place in the SpaceX Hyperloop competition. I started coding at the age of 11, learning to read English by playing with the W3 docs and dabbling on hackthissite.org , which then led me to do CTF. I later on studied Computer Science at ETH Zurich, where I fell in love with linear algebra and applied machine learning. To this day, I am generally quite curious about math- & data-heavy software challenges, especially in the Crypto domains, where I have been working on MEV and yield farming in the past. I often find myself learning about history and state theory, love good food, and am passionate about Brazilian jiu-jitsu.
We're hiring, ping me if you're interested! Working at the intersection between AI, data infrastructure and blockchain. Raised a $2M pre-seed round led by one of the leading data-VCs globally, based in Silicon Valley.
I started The AI Company with the goal of building self-learning AI systems. It then resulted as my legal entity for a number of failed companies. My biggest challenge lies in trying to build technologically innovative systems, but combining them with market needs is not always trivial. This is certainly what keeps me going though, and I'm learning (and getting closer!) with every additional attempt. - Founder and one-man-army: Scoped a 3-year pilot project and worked with a $200M turnover manufacturing facility to improve their control processes & on-site implementation of a predictive maintenance pipeline using SotA deep-learning based multivariate autoregressive models. This was tested and deployed using Docker and OPC on a LAN after consulting the factory on how to modernize their data pipelines and systems. We were building reinforcement-learning based controllers. Failed due to being too early. - Co-Founder / CTO at a medical healthcare company: Received CHF1.2M in grants and funding and hire a team of 4 backend engineers, data scientists and machine learning engineers. Failed due to co-founder issues. - Founder: GasStation, building a derivative instrument for transaction fees on the Ethereum blockchain. Failed due to missing market demand. Other occasional gigs I took since I need to pay rent (next to portfolio keeping me alive): - Lead distributed engineer at LlamaNodes: Assist in horizontal scaling, latency and throughput improvement efforts with a distributed RPC provider (https://llamanodes.com). Implemented globally distributed payment infrastructure and infrastructure to accept payment by Ethereum. - Independent contractor: Consult top-tier VCs and Family offices in transactions with a transaction volume of >$200M.
Acquisition offer by Galaxy Digital. Three pivots, starting from (1) on-chain bonds on Solana, to (2) one of the first 50 projects in Solana doing yield-aggregation with great traction until the Luna crash, to (3) a revenue-first EVM MEV project (High Frequency Trading).
Worked under Ping Li, deputy head of research at Baidu, on research-challenges to bring deep models to search engines at scale. Work included reproducing and experimenting to improve multimodal language-vision transformers (ViLBERT), and work on binary hash embeddings to scale up deep embeddings to trillions of documents using SimCLR producing binary embedding for log-search. I terminated my internship early due to my decision not to pursue a PhD in the field.
Accepted by the Google AI Impact Challenge with a ~$1M prize for our AI-heavy product benefitting marginalized communities, co-judged by Demis Hassabis, Jeff Dean, Sundar Pichai, and co. Today, Skilllab is the leading software platform for government career services. Acting as advisor since I left. Responsible for all things data & data science. Using LLM models since in 2018 combined with matrix factorization techniques. Led some product-oriented team efforts. Researched, designed and productionized the recommender engine scaling to millions of users.