Godwin-88/afriea

EA advisor for Africa # AfriEA Advisor **On-Device Enterprise Architecture & Business Process Advisory for Africa** *ADTC 2026 Submission · Corporate/Enterprise Domain* AfriEA Advisor is a fully on-device Enterprise Architecture (EA) advisory and Business Process Modelling (BPMN) platform that runs on commodity laptops common across Africa — 8 GB RAM, integrated GPU, no reliable internet required. It combines a quantized on-device LLM (Qwen2.5-3B Q4_K_M) with a local knowledge graph of enterprise capabilities, deterministic BPMN 2.0 process generation, and deep multilingual support across 12 African languages. --- ## Quick Start ### Prerequisites - Python 3.11+ - Node.js 20+ (only needed to build the frontend) - ~2.5 GB of free RAM during inference - ~2 GB free disk space for the model weights ### 1. Download the model (once, requires internet) ```bash ./scripts/setup.sh ``` This downloads `qwen2.5-3b-instruct-q4_k_m.gguf` (~1.9 GB) into `models/` and verifies its SHA-256. After download, the platform runs fully offline. ### 2. Set up the Python backend ```bash python -m venv .venv source .venv/bin/activate pip install -r backend/requirements.txt ``` ### 3. Build the knowledge graph database ```bash python convert.py --input seedgraph.cypher --output data/afriea.db --overwrite ``` This converts the Neo4j Cypher seed into the normalized SQLite schema (domains, subdomains, capabilities, standards, trends, epics, features, and all relationship tables) at `data/afriea.db`. ### 4. Build the frontend ```bash cd frontend npm install npm run build ``` ### 5. Start the platform ```bash ./start.sh ``` This brings up the FastAPI backend (serving the static React build) on `localhost`. --- ## Docker (One-Shot) The entire platform — backend, frontend build, knowledge-graph DB, and LLM runtime — can be pulled and started from a single command. The image builds the Next.js static export and installs the FastAPI/llama.cpp backend; on first run it auto-generates `data/a …