techboiafrica/natna-ai

Offline Education AI for Tigray & Leapfrog Regions. Bilingual (English/Tigrinya). Runs on one laptop, reachable from phones via local WiFi — tested at 10-15 concurrent users. Tigrinya LLM in development with Mekelle University & MiT. Domains: medicine, agriculture, mental health, college, programming. # NATNA AI **Offline Education AI for Tigray and Leapfrog Regions** NATNA is a self-contained AI education platform that runs entirely offline. It combines large language models, a 142,000+ entry Tigrinya-English dictionary, curated Wikipedia knowledge bases, and domain-specific knowledge cards for medicine, agriculture, mental health, and more. Built for communities with limited or no internet access. ## Quick Start ```bash git clone github.com cd natna-ai python installer/natna_setup.py ``` The installer automatically: 1. Downloads knowledge databases (~6 GB) from Hugging Face 2. Installs Ollama (if not already installed) 3. Lets you choose which AI models to pull 4. Verifies everything with SHA-256 checksums Then launch: ```bash # macOS ./Launch_NATNA_all_OS/NATNA_Launch_Mac.command # Linux ./Launch_NATNA_all_OS/NATNA_Launch_Linux.sh # Windows Launch_NATNA_all_OS\NATNA_Launch_Windows.bat ``` ## What Gets Downloaded | Component | Size | Source | |-----------|------|--------| | Knowledge databases | ~6 GB | Hugging Face (auto) | | AI models (your choice) | 0.4–65 GB | Ollama registry (auto) | | Ollama runtime | 29 MB–5 GB | ollama.com (auto) | **Default install (recommended models):** ~16 GB total **Full install (all models):** ~71 GB total ## Features - **Bilingual AI** — English and Tigrinya (ትግርኛ), with 142K+ hand-cleaned translation entries - **Domain intelligence** — Specialized responses for medicine, agriculture, mental health, education, programming, and more - **Wikipedia knowledge** — 4.4 GB curated Wikipedia database for context-enriched answers - **Knowledge cards** — Structured medical, agricultural, and mental health reference data - **Multiple AI models** — Switch between models in the UI. From 360M parameters to 14B+ - **Fully offline** — Everything runs locally after setup. No cloud, no API keys, no subscriptions - **Cross-platform** — macOS (Intel & Apple Silicon), Linux, Windows ## Tigrinya La …