Luvira-LLC/vigilant-gemma

AI-powered safety net for African health facilities that links newborns to maternal HIV records and flags hidden adherence risks. # VIGILANT — Gemma 4 Good Hackathon **Local-First AI for Infant HIV Risk Detection** Built with Gemma 4 via Ollama | Zero cloud dependency | NDPA 2023 compliant ## Quick Start 1. Read `SUBMISSION_GUIDE.md` — full step-by-step checklist 2. Upload `VIGILANT_Gemma4_Kaggle_Notebook.ipynb` to Kaggle (GPU enabled) 3. Paste `VIGILANT_Gemma4_Writeup.md` into Kaggle Writeup form 4. Record video using `VIDEO_SCRIPT.md` 5. Submit before **May 18, 2026** ## Files (7 total) | # | File | Purpose | | --- | ----------------------------------------- | ------------------------------------------------- | | 1 | `VIGILANT_Gemma4_Kaggle_Notebook.ipynb` | **The code** — self-contained Kaggle notebook | | 2 | `VIGILANT_Gemma4_Writeup.md` | **The writeup** — paste into Kaggle (≤1500 words) | | 3 | `VIDEO_SCRIPT.md` | **Video script** — 3-min demo guide | | 4 | `SUBMISSION_GUIDE.md` | **Submission checklist** — what to upload where | | 5 | `VIGILANT_Gemma4_Proposal.md` | Design doc — read for context | | 6 | `VIGILANT_Gemma4_Implementation_Guide.md` | Technical blueprint — read for context | | 7 | `README.md` | This file | ## Architecture ``` Infant Record → Forensic Linker → Mother Match (Top 3) ↓ Mother's Notes → Adherence Miner (Gemma 4) → Risk Indicators ↓ Hallucination Firewall ↓ VL + Adherence → Protocol Guardian → Risk Level + Drug Regimen ↓ Role-Based Filter → Nurse / Specialist View ↓ SHA-256 Audit Chain → NDPA Compliance Log ``` ## Target Tracks - **Health & Sciences** (Impact Track) — $10,000 - **Ollama** (Special Technology Track) — $10,000 - **Main Track** — up to $50,000