SUBMISSION_NOTES.md

Submission Notes

Track

Backyard AI.

Reason: the problem came from a real architecture student / early-career architect, and the judging story should focus on whether the app helped her prepare site-analysis sheet material faster.

Strongest Demo Story

Architecture students often receive a site from faculty, Google Earth, CAD, or rough map marking. Before design begins, they collect scattered site conditions: climate, sun, wind, access, roads, water, vegetation, terrain, soil, existing features, and local activity. Site Intelligence Studio turns that first boundary into a board pack with diagrams, captions, evidence, and a site-visit checklist.

What To Show In 60-90 Seconds

1. Manual pain: site-analysis data is scattered and then redrawn into sheets.
2. Click Use Chorwad sample site or upload a real KML/DXF.
3. Generate board pack.
4. Show board PNG.
5. Show context diagram with OSM geometry, climate chart, and sun/wind diagram.
6. Show Evidence Table with source/confidence/limits.
7. Show Satellite/Earth Reference links and safety note.
8. Show site-visit checklist.
9. Show Small Model Brief. If live small-model inference is not configured, explain the fallback honestly: facts are computed/retrieved; a <=4B model can only rewrite evidence into captions and checklist language.
10. End with friend validation quote once collected.

Social Post Draft

I built Site Intelligence Studio for the Hugging Face Build Small Hackathon.

It helps architecture students turn a site boundary from a map, KML/KMZ, GeoJSON, or CAD/DXF into a preliminary site-analysis board pack: climate chart, sun/wind notes, context map, terrain/soil caution, evidence table, and site-visit checklist.

The important part is what it refuses to do: no legal boundary claims, no foundation recommendations, no fake culture/demographic inference, and no replacement for a site visit. The app separates computed facts, public data, uploaded-file facts, user notes, and verification items.

Built with Gradio for the Backyard AI track.

Blog / Field Notes Outline

1. The friend's original problem.
2. Why generic AI report generation is not enough for architecture students.
3. Boundary-first product design.
4. Data sources and their limits.
5. What the app automates and what it intentionally leaves to humans.
6. Safety rules for soil, foundation, legal boundary, and culture.
7. What friend validation showed.
8. What a future production version would need.

Submission Readiness Gate

Do not call the project final until:

• the Space loads
• sample site generates a board
• drawn polygon still works
• pin-radius still works
• KML/KMZ or GeoJSON upload works
• DXF candidate + anchor workflow works
• board PNG/PDF download works
• evidence table is readable
• tests pass
• friend validation is collected
• demo video and social post are ready