Second Opinion
Second Opinion is a multi-agent healthcare AI tool that turns a patient's chart into a cited, fact-checked appointment brief designed for under-researched conditions starting with adenomyosis, a disease that takes 8 year
Core trick
Runs three specialist agents in parallel over a fixed guideline corpus, then a verifier flags every uncited claim before the brief ships.
Tags
Description
Second Opinion is a multi-agent healthcare AI tool that turns a patient's chart into a cited, fact-checked appointment brief designed for under-researched conditions starting with adenomyosis, a disease that takes 8 years on average to diagnose. A patient drops in their full chart. Five Opus 4.7 agents run a structured consultation: three specialists fire in parallel (OB-GYN, Reproductive Endocrinology, Chronic Pain) consulting from a 32K-token corpus of 2026 guidelines, active clinical trials, and care pathways. A
synthesizer composes one cohesive brief. A verifier fact-checks every claim against the corpus and flags anything uncited — green badge = every claim sourced.
The brief is research, not advice. Patient charts never persist on disk. The doctor's chair stays the doctor's chair. Output speaks 5 languages.
Built solo in 5 days on Opus 4.7's 1M context window with prompt caching. Adenomyosis is the start; the same pipeline answers fibroids,
endometriosis, Hashimoto's, long COVID, lupus, POTS — any condition that takes too long, gets dismissed too often, and has a literature too
dense for one appointment to hold. The corpus changes; the architecture doesn't.
Transparency: "Sandra" is an anonymized real patient profile (with consent). "Maya" and "Helena" are hand-authored composite cases. All five
specialists are Opus 4.7 agents not human doctors. Care centers and citations are sourced from public medical literature.
The linked repo returns 404, so it was deleted or made private.