Research notes
Methods, failure studies, and open questions.
Working notes from our research on ontology, source structure, changing evidence, and grounded medical knowledge systems.
The Bitter Lesson
What the Bitter Lesson means for medical AI, from frontier model performance to auditable systems and better evaluation.
The Neurosymbolic Advantage: Why Structured Data Beats Bigger Models
ESMO guidelines aren't just documents — they're a clinical ontology. Here's why using them as a symbolic scaffold for LLM-based retrieval outperforms pure neural approaches.
What Happens When Your AI Recommends an FDA-Rejected Drug
In our study, Perplexity recommended a drug combination rejected twice by the FDA. Claude suggested a withdrawn medication. This isn't just an accuracy problem — it's a patient safety problem.
Why We Built Grounded Retrieval
The origin story of Grounded Retrieval — built by a practicing medical oncologist who needed better answers than frontier AI models could provide.