Physicist · Entrepreneur · Builder
From CERN particle detectors to Hopfield neural networks to AI agents — I build software at the intersection of science, technology, and business. Currently developing LLM-powered email-based services.
Z80 assembler on a ZX Spectrum. A degree in Physics. Particle detector simulations at CERN. A PhD on stochastic neural networks at ETH Zürich. Software companies in finance, manufacturing, and AI. The common thread: computation that ships. Currently building LLM-powered agents delivered entirely through email.
Italian mother tongue · Highly proficient English · Good German · Basic French and Spanish
PhD in computational neuroscience, ETH Zürich's Institute for Neuroinformatics. Publications on Nature Biotechnology, Biotechnology Progress, Biological Cybernetics. Earlier at ETH: robotics, gene and biochemical network modelling. Before that, particle detector work at CERN.
Founded and led several software companies: derivatives pricing algorithms, IoT architecture for medical devices, machine-learning-based text-analytics engine, e-commerce platform for 3D-printing bureaus. In between: financial risk consulting at Oliver Wyman.
Now: LLM-powered services delivered entirely through email. Silex is the infrastructure platform. Aleik is the first product running on it.
Platform · Infrastructure
The platform for LLM-powered email-based services.
Silex lets you build intelligent, conversational services that run entirely through email — no app, no login, no installation required. It handles LLM orchestration, conversation memory, human approval workflows, and logging, on proven infrastructure built with Erlang, Elixir, and PostgreSQL.
botwerk.com →Product · AI Research Writing
Science finds the story. You tell it. We write it.
Aleik surfaces new scientific research as weekly email digests, lets writers choose a story angle by reply, and delivers a publication-ready article — fully sourced, SEO-optimised, written in the subscriber's own voice. The entire workflow runs through email. Built on Silex.
aleik.com →If minds are not the exclusive property of brains, the carbon/silicon line stops being the interesting one. On Rouleau and Levin’s continuity argument, the gap Be Right Back is really about, and what testing can and cannot see.
If an AI persona could produce the same behavioural variation as a human tester, would the practical advantage of human testing disappear? On Lem’s Electronic Bard, Searle’s Chinese Room, and the bet underneath the methodology.
Eight methods for testing AI agents — from persona-based and contract testing to mutation, regression, and human red-teaming — organised into three tiers of evaluation difficulty.
The model didn’t change its mind — it has no mind to change. On sampling, anthropomorphism, and treating system prompts as code.
Why I converted most of my AI code from Python to Elixir — preemptive scheduling, supervision trees, and clean retry semantics for LLM-driven systems.
Prompt engineering is often seen as an art form. A methodology for systematically capturing the tacit knowledge gained during iterative refinement.
Every year someone announces the death of email. What if its universality is exactly what makes it the most interesting deployment surface for AI agents?
A Hugging Face study argues the risks of fully autonomous AI agents significantly outweigh the benefits — and makes the case for keeping humans in the loop.