Sanchai - Zero-Hallucination Nepali Clinical Health Ledger & Multimodal Ingestion Engine
About the project
What is Sanchai?
In Nepal, healthcare documentation is fragmented. Patients carry fragile paper discharge summaries, handwritten doctor prescriptions, and lab printouts written in mixed Nepali, English, and Romanized slang (“jwaro aako”, “khoki”, “cetamol 500”). When patients visit different clinics or emergency rooms, doctors have no immediate medical history, leading to repeated lab tests, misdiagnoses, and dangerous drug interactions.
Sanchai (सञ्चै) is an offline-first, bilingual longitudinal electronic health record (EHR) built specifically for Nepal’s healthcare ecosystem. It pairs a deterministic Nepali Clinical Lexicon (260+ curated concepts with ICD-11 linkages) with Google Gemma 4 guardrails to deliver 100% verifiable clinical data without hallucination.
Key Features
- Multimodal Clinical Intake: Upload photo scans, PDFs, or type doctor notes in Nepali, English, or mixed slang. Gemma 4 OCR and pipeline extracts entities, cross-references them against our clinical lexicon, and links them to formal WHO ICD-11 codes.
- Zero-Hallucination Assertion Verification: Clinical concepts (like “Fever Negated” vs “Fever Present”) are verified deterministically. Gemma 4 is used strictly for candidate extraction and clinical translation; no ungrounded medical claims can ever enter the authoritative record.
- Emergency QR Triage (Works 100% Offline): Patients carry a tamper-proof QR code on their phone or printed card. In an emergency, any paramedic or doctor scans it with a phone camera or webcam to view blood group, critical drug allergies (e.g. Penicillin), active conditions, and emergency contacts in under 1 second without internet access.
- SanchAI Bilingual Health Assistant: An in-app clinical chatbot that understands the patient’s entire longitudinal history. Patients can ask questions in Nepali or English, receive pre-visit consultation summaries, or ask for clear explanations of lab results and prescriptions.
- Print-Ready A4 Doctor Summaries: Generates clean, standardized A4 PDF clinical summaries and HL7 FHIR R4 interoperability bundles with a single click.
-
Dual-Tier Offline Fallback: Runs in cloud mode or completely offline using local quantized Gemma 4 (
gemma4:e2b-it-qatvia Ollama) on low-resource hardware in rural health posts.
Real-World Impact
Over 30 million people in Nepal face fragmented care across private clinics and government hospitals. Sanchai bridges the language and technology gap with an open-source, standards-compliant system that works in the most remote health posts of the Himalayas.
Built with
AI tools
Prize Categories Entered
- Submitted
Best Use of Gemma 4
- Event prizeSubmitted
Best use of open source AI for Real-World Impact
- Event prizeSubmitted
Best Nepali / Local-Language AI