Knowledge shorts twice: Once in the tools. Once in the humans.
I build tools for complex knowledge and test whether they work. From user research through working code — end to end.
The problem is always the same: complex information that people can't use well enough. Sometimes that's a researcher drowning in literature. Sometimes it's a checkout flow losing customers. Sometimes it's an organization sitting on data nobody can find.
Full e-commerce rebuild for Rome's esoteric bookshop.
A computable UX knowledge system — HCI research encoded as a graph with every citation verified against its source, exposed to Claude Code through an MCP server so evidence lands in a developer's context window exactly when they need it.
Barcode to catalog in under 10 seconds. A metadata aggregation platform for independent bookshops — scan, enrich from five sources, publish to WooCommerce.
Multi-agent research engine that takes an impossible hypothesis, stress-tests it against real academic literature, and produces papers that read like hard science.
The throughline has always been one thing: rigorous evaluation of how humans interact with complex information systems, and building better ones.
It started with classical philology — close reading as a contemplative discipline, sitting with a text until it reveals what it actually says rather than what you wanted it to say. My MA thesis traced the arc from Roberto Busa's Index Thomisticus to modern computational text analysis.
Then HCI research at Sapienza and Cologne, funded by a Marie Curie fellowship: are digital scholarly editions actually serving their users? (Usually not — scholars preferred the PDF.) Three papers, lectures from Berlin to Tokyo to Buenos Aires, a PhD formalizing evaluation frameworks for knowledge tools.
Then eight years at Baymard Institute — the same question applied at scale. 700+ usability heuristics. Fortune 500 checkout flows. Empirical proof, over and over, that the tools fail and the humans fail harder.
Impossible Papers is where it all converges: a system that attacks both failure modes simultaneously. The tooling problem (multi-agent pipeline, automated literature, adversarial review) and the human problem (the system is the furnace, not the alchemist — human judgment stays exactly where it belongs, and gets removed everywhere it breaks things).
If you're working on something where the knowledge infrastructure is the bottleneck — whether that's a research lab, a scientific competition, or a business drowning in its own data — I'd like to hear about it.
Get in touch