thepreakerebi/fundi

Fundi — offline on-device biomedical-equipment maintenance copilot. Africa Deep Tech Challenge 2026 (The Laptop LLM Challenge). # Fundi 🛠️🩺 **An offline, on-device biomedical-equipment maintenance copilot for low-resource clinics.** Runs entirely on a clinic's existing 8 GB laptop — no GPU, no internet. A technician describes a symptom, pastes an error code, or feeds a device log, and Fundi returns a **likely failure mode, offline checks to run first, and an escalate / order-part / resolve-on-site decision** — every answer cited to real-world failure reports. > Entry to the **Africa Deep Tech Challenge 2026 — "The Laptop LLM Challenge."** > 100% offline · llama.cpp + GGUF · ~1.9 GB peak RAM · English / Kiswahili / Yorùbá / Igbo --- ## The problem Across sub-Saharan Africa, **38–70% of medical and lab equipment is non-functional at any time**, and there are **fewer than 10 biomedical engineers per million people** (vs 100+ in high-income countries). When a ventilator, centrifuge, or analyzer fails in a district clinic, there's often no on-site engineer, no manufacturer support, and no internet to reach one. The knowledge to fix it exists — just not where and when it's needed. ## What it does - **Plain-language triage** for the equipment a technician actually maintains — ventilators, infusion pumps, patient monitors, defibrillators, anesthesia machines, sterilizers, lab analyzers. - **Structured, cited answers**: `Likely cause → Check first → Action`, grounded in real failure reports. Click any source to read the actual report text behind it. - **Fully offline** on an 8 GB CPU-only laptop. - **Works in English, Kiswahili, Yorùbá, and Igbo** via an offline translate-bridge. ## How it works ``` symptom / error / log │ (optional MT: sw/yo/ig → en) ← MADLAD-400, offline ▼ BM25 retrieval over the knowledge base ← FDA MAUDE failure cards + WHO guidance ▼ Qwen2.5-1.5B (GGUF Q4_K_M) via llama.cpp → cited Likely cause / Check first / Action │ (optional MT: en → sw/yo/ig) ``` - **Core model:** Qwen2.5-1.5B-Instruct (Apache-2.0), quantized to **GGUF Q4_K_M**, served via **llama.cpp** — …