Med Trace AI

MedTrace AI supports doctors, nurses, and general practitioners where specialists and clean electronic records are scarce. It accepts the messy reality of care—handwritten prescriptions, scanned labs, vitals, notes, transcripts, and images—and extracts medical facts with links back to sources. Those facts feed a temporal patient knowledge graph, so clinicians can ask questions like “what changed before this symptom?” instead of piecing the story from memory alone. When a clinician runs the Cause Explorer, the system retrieves patient-grounded context first, then uses a fine-tuned open medical model (with efficient tuning such as LoRA/QLoRA) to output possible causes to investigate, supporting evidence, missing information, red flags, and next questions—always framed for review, never as a definitive diagnosis or treatment plan. The hackathon MVP demonstrates dashboard, uploads, extraction review, timeline, graph view, Cause Explorer, safety messaging, and an AMD-focused benchmark (latency, VRAM, structured output). Open models and AMD optimization matter because they move capable, auditable clinical support closer to rural clinics and district hospitals—where one missed detail can change a life.