Logo Lanfrica

gbohigbaradc/mediassist_submission

Domain:

healthcarenatural language processing

Record type:

softwareproject
Creator:
gbo
Host:
Offline AI medical assistant for rural Africa - Africa Deep Tech Challenge 2026 # ⚕️ MediAssist — Offline Medical Assistant ### Africa Deep Tech Challenge 2026 | Healthcare & Medical Domain > AI-powered medical triage and guidance that runs 100% offline on an 8GB laptop. > No cloud. No GPU. No internet required. --- ## Problem Rural communities across Nigeria and Africa face a critical gap: qualified health workers are scarce, and when patients need medical guidance, the nearest clinic may be hours away. Cloud-based AI health tools require stable internet — a luxury unavailable in most rural settings. **MediAssist solves this by bringing intelligent medical assistance directly to the device.** --- ## Solution A lightweight offline medical chatbot built on **Qwen2.5 3B** running via **Ollama**, served through a **Python Flask** backend with a clean HTML/JS frontend. It supports: - Symptom triage and referral decisions - Medication guidance and dosing - Maternal and child health - General primary care Q&A - 5 languages: English, Nigerian Pidgin, Hausa, Yoruba, Igbo --- ## ADTC Compliance | Requirement | Status | |-------------|--------| | Runs 100% offline | ✅ | | No cloud dependency | ✅ | | No discrete GPU | ✅ | | Fits in 8GB RAM | ✅ (~2.5GB RAM usage) | | Ubuntu 22.04 compatible | ✅ | | Healthcare domain | ✅ | **Estimated scores:** - Sacc: High — medically safe, structured, multilingual responses - Sperf: ~15–20 TPS on i5 10th–13th gen (num_thread=4, num_ctx=2048) - Seff: ~65/100 — Peak RAM ~2.5GB out of 7GB budget --- ## Tech Stack | Component | Choice | Why | |-----------|--------|-----| | LLM | Qwen2.5 3B | Best multilingual quality + lowest RAM on 8GB laptop | | Runtime | Ollama | Simple local model server, cross-platform | | Backend | Python Flask | Lightweight, no overhead | | Frontend | Vanilla HTML/CSS/JS | Zero dependencies, works in any browser | | RAM usage | ~2.5GB | Leaves headroom for OS and app | --- ## Quick Start ### 1. Install Ollama ```bash # Ubuntu curl -fsSL ollama.com | sh # Windows: …