Sangam : Polypharmacy Safety Council

Sangam is a multi-agent pharmaceutical safety system built on Band AI's infrastructure. Six specialist agents — Intake, PatientProfile, StructuralBio, PKPD, EvidenceRAG, and ComplianceGuard — coordinate via @mention routing inside a shared Band room to screen any combination of allopathic drugs and Ayurvedic herbs for dangerous interactions. The problem it solves is real and largely ignored. India has the world's highest rate of concurrent allopathic and traditional medicine use. Up to 70% of patients never disclose herbal supplement use to their doctor. No clinical tool exists to screen these combinations systematically — every major drug interaction checker screens against the Western pharmacopoeia only, with zero Ayurvedic herb coverage. Each agent handles a distinct analytical layer. The Intake Agent parses free-text patient queries and fetches compound data from PubChem. PatientProfile computes a personalized clearance modifier based on CYP2C9/3A4 genotype, renal function, and age. StructuralBio queries a curated molecular docking database of 26 drug-herb pairs across six enzyme targets — CYP1A2, CYP2C9, CYP2C19, CYP3A4, P-gp, and OCT — returning binding affinity in ΔG kcal/mol. The PKPD Agent runs a one-compartment pharmacokinetic model and computes AUC percentage change with a full 48-hour concentration curve. EvidenceRAG retrieves supporting findings from a curated corpus of 70 peer-reviewed studies and traditional pharmacology texts. ComplianceGuard synthesizes all five upstream reports and issues a RED, YELLOW, or GREEN verdict with confidence score, clinical action, and regulatory disclaimer. The system also includes a fast deterministic combination screener at /api/interactions/screen that returns pairwise risk verdicts in milliseconds without an LLM call — built for point-of-care use. The stack is FastAPI backend, React + Vite frontend with WebSocket streaming, ChromaDB vector index, DeepSeek LLM, Docker Compose deployment, and GitHub Actions CI.