WHY: In Indonesia, patient medical records do not follow patients between facilities. A rural clinic (Puskesmas) writes notes on paper, a district hospital (RSUD) receives referrals with missing clinical history, and patients with low health literacy cannot recall their medication history. Standard AI text search fails because it stores every mention separately, clinically dangerous contradictions occur when active and stopped medications are displayed side-by-side.
WHAT: Engramic is a stateful patient record workspace. It processes bilingual clinical notes and PDFs, extracting key entities (conditions, medications, allergies, labs) and mapping them into an updateable, unified patient graph. It ensures that when a medication is discontinued or a condition is resolved, the status is updated globally instead of generating duplicate, conflicting records.
HOW: We built a full-stack Python clinical system. A clinician inserts text or drops a clinical PDF. Engramic runs an AI-first extraction pipeline (with local OCR fallback via Tesseract). Extracted facts are displayed in a clinician review workspace where the doctor can edit, include (Sertakan), or reject (Tolak) each point. Approved facts are merged into an SQLite graph store using a fuzzy string matching algorithm (rapidfuzz) to automatically resolve synonyms and track historical provenance.
Log in or sign up for Devpost to join the conversation.