Tunzeki/adtc-hackathon

Submission for the Africa Deep Tech Challenge 2026 # Offline Clinical Decision Support & Triage for Primary Healthcare (ADTC 2026) An offline-first, AI-assisted clinical decision support system built for the **Africa Deep Tech Challenge (ADTC 2026)**. Powered by **Gemma 4 E2B** (`Q8_0` GGUF) and executed via **`llama.cpp`**, this tool provides real-time clinical triage, risk evaluation, and disposition recommendations for Community Health Extension Workers (CHEWs) operating in primary healthcare centers (PHCs) across Nigeria and sub-Saharan Africa—100% offline without internet access or GPU hardware. --- ## 💡 The Problem In Nigeria, Community Health Extension Workers (CHEWs), Junior CHEWs (JCHEWs), and other non-physician clinicians (NPCs) deliver front-line primary care at Level 1 (community) and Level 2 (PHC clinic) facilities. In rural areas where medical doctors are rarely present, NPCs must make rapid, high-stakes clinical decisions from memory: * **Manage on an outpatient basis** * **Observe** for a specific duration * **Admit** for local treatment * **Refer immediately** to a specialist or secondary facility A triage misjudgment—such as admitting a pediatric patient with severe malaria and altered consciousness at a Level-2 clinic instead of arranging emergency referral—can prove fatal. This project equips health workers with a reliable, local AI clinical assistant to ensure evidence-based, safe, and timely triage at the point of care. For more details on the problem statement, design decisions, operational constraint, and benchmark, please see REPORT.md --- ## Local Model Installation & Testing Submission First, install ollama with ```bash curl -fsSL ollama.com | sh ``` Confirm a successful installation with ```bash ollama --version ``` Next, install huggingface_hub ```bash pip install huggingface_hub ``` Install gemma4-e2b-it from huggingface ```bash llama cli -hf ggml-org/gemma-4-E2B-it-GGUF ``` Start the server ```bash llama server ``` Alternatively, find out …