bugindacodeQ/tibaedge-adtc-2026

TibaEdge — a safety-first, offline healthcare AI assistant for frontline African health workers. ADTC 2026 submission. # TibaEdge TibaEdge is a safety-first, offline language-model concept for trained frontline health workers in African settings where connectivity, compute, and access to reference material may be constrained. This repository is the team's entry for the **Africa Deep Tech Challenge 2026 — Healthcare & Medical domain**. > [!IMPORTANT] > TibaEdge is an early prototype and research submission. It is not a medical > device, has not been clinically validated, and must not be used to diagnose, > prescribe, or replace qualified medical judgement. ## Submission baseline - **Runtime:** `llama.cpp` - **Model:** Qwen2.5-1.5B-Instruct - **Format:** GGUF Q4_K_M - **Target:** 4 vCPU, 8 GB RAM, integrated graphics, Ubuntu 22.04 - **Connectivity:** model inference is fully offline - **Primary language:** English The current package is a reproducible general-model baseline. Domain adaptation, retrieval over approved clinical references, safety evaluation, and review by qualified health professionals remain future work and are not represented as completed features. ## Repository structure ```text . ├── metadata.json # ADTC submission and model metadata ├── download_model.sh # credential-free, idempotent GGUF download ├── REPORT.md # technical report ├── model/ # downloaded weights; ignored by git └── assets/ # project artwork ``` ## Run locally Install a current `llama.cpp` build so `llama-cli` and `llama-bench` are on the `PATH`, then run: ```bash bash download_model.sh llama-cli -m model/qwen2.5-1.5b-instruct-q4_k_m.gguf -cnv ``` No network access is required after `download_model.sh` completes. ## Run the ADTC participant profiler The official profiler requires Python 3.11+ and `llama-bench`: ```bash python -m pip install "git+github.com" adtc-profiler run \ --submission . \ --mode participant \ --output submission.json \ --skip-accuracy ``` `submission.json` is deliberat …