# Wagner Solutions

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Sebastian Wagner](https://cerebralvalley.ai/u/SebaWag)
- **GitHub:** https://github.com/SebaWag/expediente
- **Demo video:** https://youtu.be/jGUUaP40RaI
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2477

Chile publishes five years of public procurement — 10,302,746 records across 240 archives — and
also publishes its audit findings. The data exists. What does not exist is the ability to connect
it and prove it.

Expediente is a verifiable evidence engine. You ask a question in plain Spanish; the agent writes a
plan naming tools that exist, picks one per turn, and walks a graph of 6,567,657 purchase orders.
It never writes SQL and never touches the graph directly: it chooses among thirteen bounded verbs,
and each one compiles to a traversal with limits. That is why eight steps are enough, and why the
answer is reproducible tomorrow.

Every claim carries the order, the amount, the date and the official record with its sha256. A
separate adversarial gate deletes any claim it cannot trace to a document. The card in the demo does
not say "eight orders" — it says two of eight, that they are the supplier's, that they are 81.52% of
the spending, and that both have a document. What the system left out is stated on screen, not in a
footnote.

Today: 6,567,657 orders, 145,600 suppliers, 2,065 buyers (100% of the corpus), 565 signals, 4,455
documents, 86,012,436 triples. A run costs 45,905 tokens and 44.7 seconds, 98% of it model time.

## More from Meta Global AI Developer Hackathon

- [Heywizzle](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2474)
- [Uyari](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2475)
- [NoCap](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2476)
- [meghna tighare](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2478)
- [egan](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2479)
- [Airlock](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2480)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2477. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt