khelifamoussa/AEGIS-OMEGA-AFRICA

Offline-first autonomous AI crisis decision-support system for Africa, powered by a deterministic safety core with optional local Qwen3 refinement via llama.cpp. # AEGIS OMEGA AFRICA V8 **Offline, deterministic crisis decision-support for connectivity-constrained environments** AEGIS OMEGA AFRICA V8 is a competition prototype designed to produce an immediate, auditable crisis-response plan on a standard laptop without depending on cloud services. Its deterministic decision core prioritizes safety constraints, rejects invalid scenarios, prevents resource over-allocation, records key unknowns, and can optionally request a local Qwen3 model refinement through llama.cpp. ## Why AEGIS During floods, wildfires, infrastructure failures, remote-community emergencies, and communications outages, cloud-dependent AI may become unavailable exactly when decisions are most urgent. AEGIS follows a **safe-first architecture**: 1. **Immediate deterministic decision** --- produces the safe baseline without waiting for an LLM. 2. **Constraint validation** --- checks scenario consistency and resource limits. 3. **Optional local-AI refinement** --- uses a local Qwen3 model only when available. 4. **Safety gate / fallback** --- if local AI fails, times out, or is unavailable, the verified deterministic plan is retained. 5. **Audit output** --- saves decision metadata and timing to JSON. ## Competition Track Primary fit: **Autonomous AI Agents / offline local orchestration and privacy-focused decision support**. The project is designed around the Africa Deep Tech Challenge requirement for an end-to-end, on-device language-model solution that can operate without cloud dependencies on the reference laptop class. ## Core Features - Fully local crisis decision core - No cloud dependency for the deterministic path - Fast immediate decision path - Deterministic safety constraints - Resource-allocation guardrails - Invalid-input rejection - Explicit critical dependencies - Explicit key unknowns instead of invented facts - Optional local Qwen3 refinement through llama.cpp - Safe fallback when the model is unavailable or …