What building it taught me
I build with AI, and the craft is knowing where to slow down. Prototyping fast let me skip two decisions I paid for later, the stack and security. I retrofitted multi-tenancykeeping every user's data cleanly separated so no one can ever see anyone else's. because speed had skipped it, so now I decide who I’m building for and what they need first.
Researchers need their data secure and every result auditable, and that requirement drove the whole design. Every categorization traces back to its origin text, the pipeline is deterministicit gives the exact same answer every run, so a person can always check the machine., and I chose Hetzner for EU data-protection.
Where it stands
It’s live at syndesmos.ai, and now being shaped into an NIH R21 grant with UW-Madison faculty for drug-safety signal detection.