Hera is a multi-agent medical research assistant, built on Google's Agent Development Kit (ADK), that turns a patient story into a cited, patient-readable research brief in minutes.
A user pastes a free-text case — e.g. "67-year-old with vision loss, MacTel, type-2 diabetes, and stage-3A CKD" — and Hera returns an Expert Research brief: the most relevant published papers (with citations), named researchers worth consulting, recruiting clinical trials, citation-grounded "questions to ask your specialist," and a personalized literature feed. Everything is grounded in real sources, streamed token-by-token, and exportable to PDF.
Under the hood, an ADK SequentialAgent chains intake → research → topics → synthesis, while a ParallelAgent fans out paper search, researcher matching, clinical-trials lookup, and Grounding-with-Google-Search concurrently — roughly 2× faster than our prior sequential path. Clinical-trials data is consumed over the Model Context Protocol via ADK's McpToolset. A dedicated, safety-constrained Gemini agent writes the "topics to discuss": it must cite a source already in the brief, never recommends doses, and falls back to a disclaimer if it can't safely ground at least three topics.
It's not a prototype - Hera already runs in production for clinicians and patients at synapsesocial.com/health-cases, backed by OpenAlex, PubMed, ClinicalTrials.gov, the NPI Registry, and Synapse's own paper corpus.
