# MaestrIA

- **Event:** [Built with Opus 4.7: a Claude Code hackathon](https://cerebralvalley.ai/e/built-with-4-7-hackathon)
- **When:** Apr 21 at 12:00 PM – Apr 27 at 2:00 AM (EDT)
- **Where:** Online
- **Placement:** Finalist
- **Team:** [Benjamin Torralbo](https://cerebralvalley.ai/u/Benja)
- **GitHub:** https://github.com/Benjxyg/MaestrIA
- **Demo video:** https://youtu.be/rkH4AjoTL5Q
- **Gallery:** https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery/194

My dad is a master carpenter in Chiloé, southern Chile. 30 years on
the job, most of them as site foreman, and 8 of those years
restoring heritage churches with the Fundación Iglesias Patrimoniales
de Chiloé. I am his apprentice and I also run the development and
business side of his company, Constructora Patrimonial Chiloé. I
built its website (constructorapatrimonialchiloe.cl) with Claude
Sonnet a few months ago. The website helps people find him. It does
nothing to help a client know whether his quote is fair, whether his
diagnosis is correct, or whether the trade showing up is the right one
for the problem. In Chile alone, 280,000 construction workers are in
the same position (INE, 2025).

MaestrIA is what I built so a regular family can get an architect-grade
diagnosis for the kind of repair an architect would never come to see.
You take a photo, describe the problem in your voice in Chilean Spanish,
and say where you live. Claude Opus 4.7 thinks live on screen, draws
boxes over the damage as it reasons, names the cause, estimates the
work in real units (half a day, two days, not "3.5 hours"), and pulls
together a committee of trades that argue on screen until they agree
on what to do. A managed agent checks the prices against Chilean
retail. A second agent writes the WhatsApp message to the chosen
maestro for you: greeting by name, problem summary, materials, ballpark
budget, time estimate, link to the full diagnosis. The client stops
arriving with "my wall got wet" and starts arriving with a brief.

The whole system was measured against my dad. He diagnosed 12 real
photos in audio, without seeing the AI. Across 9 dimensions
(pathology, material, days of work, CLP cost, severity, honesty under
uncertainty, and others), MaestrIA agrees with him 81% of the time.
Before integrating his interview into the knowledge base, the score
was 74%. Same model, same prompt, same photos. The +7 points are his.

---

Markdown version of https://cerebralvalley.ai/e/built-with-4-7-hackathon/hackathon/gallery/194. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt