dan-angila/atlas-adtc2026

Offline-first healthcare reference intelligence assistant. Cited RAG answers over local clinical documents, running entirely on 8GB commodity hardware with zero network. Rust + llama.cpp + SQLite. Africa Deep Tech Challenge 2026. # BRIX Atlas **An offline-first healthcare reference intelligence assistant that runs powerful AI entirely on commodity hardware — no cloud, no network, no compromise on privacy.** Official implementation for the **Africa Deep Tech Challenge 2026**. BRIX Atlas is **not a medical device** and does not diagnose, triage, or recommend treatment on its own authority. It answers questions grounded in clinical/health-reference documents an organization has deliberately loaded — every answer is traceable back to a specific loaded source. See ADR-0014 for the reasoning behind this vertical, and SECURITY.md for the full safety posture. The underlying engine itself remains domain-agnostic — a clinical guideline PDF is parsed exactly like any other PDF. -orange) --- ## This is not a prototype This repository is the engineering foundation of a production-grade, open-source AI platform — not a hackathon demo and not a one-off competition submission. It is engineered from day one for a long life and a large contributor base, under the constraints of the hardware most businesses actually have, not the hardware AI demos are usually built on. The long-term vision is for BRIX Atlas to become the AI engine powering the wider BRIX ecosystem. ## Why this exists Most "run AI locally" products quietly assume a GPU, or 32GB+ of RAM, or a connection to the internet for at least the heavy lifting. A clinic or health facility with unreliable or no connectivity — common across large parts of the world — can't rely on any of those, yet that is exactly where fast, trustworthy access to treatment guidelines and drug formularies matters most. BRIX Atlas is built for that reality: **Ubuntu 22.04, an Intel Core i5 (10th–12th Gen) or AMD Ryzen 5, 8GB of RAM, integrated graphics only, and zero network connectivity required, ever.** Within that envelope, the platform lets a clinician, pharmacist, or community health worker: - Chat with locally-loaded clinical reference documents — PDF, DOCX …