GiziGuard — On-Site AI Nutrition Assistant

An AI assistant for school kitchens that runs entirely on-site — no internet needed, no data leaves the building. Trained on Indonesia's official food composition database to validate nutrition automatically. Submitted to the Kaggle Gemma 4 Good Hackathon.

GitHub Kaggle

Highlights

Runs offline — no data ever leaves school premises
Understands both Bahasa Indonesia and Sundanese
Combines camera vision and language AI on affordable hardware

Problem

Kitchens under the school nutrition programme need real-time nutrition validation, but schools have unreliable internet and strict expectations that data stays on-premises. Cloud-hosted AI is neither reliable nor appropriate here.

Approach

Trained a custom AI model on Indonesian and Sundanese cooking and nutrition data, matching the pilot kitchen's language mix
Grounded the AI's answers in Indonesia's official food composition database, so it cites real numbers instead of guessing
CCTV activity recognition tracks kitchen events — receiving, prep, plating, distribution — using the existing camera system
Runs entirely on a small computer at the kitchen — no internet connection needed, no data leaves the premises
Plugs into the DPMBG system as a separate nutrition-analysis service

Stack

Gemma 4UnslothOllamaPythonHik-Connect CCTVRAGTKPI DatasetFastAPI

Outcome

Submission targeted for the Kaggle Gemma 4 Good Hackathon (deadline 2026-05-18)
Reusable on-site AI layer applicable to other government nutrition programmes