# Centurion

- **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:** [Brandon Lim](https://cerebralvalley.ai/u/brandonlim)
- **GitHub:** https://github.com/prodbrandon/centurion
- **Demo video:** https://drive.google.com/drive/folders/1Q-rpun4OfblXQ8tK00Io0HTTpmlOedUK?usp=sharing
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5145

Centurion's promise to a brand is trust: every number on screen can be traced to an order, a catalog fact or a source. Muse Spark is what turns that data into something a busy merchant can act on in seconds, without breaking the trust.

1. Meta AI is one of the assistants shoppers use. Testing what Meta's own model says about a store's products is direct evidence of what shoppers hear. In our live runs Muse Spark declined to guess about a store it didn't know instead of inventing prices or specs, exactly the behavior a brand wants from the model that answers its customers, and the behavior our analyst depends on.

2. Grounded, structured output. Muse Spark follows a strict "answer only from this data and quote the numbers" instruction and returns valid JSON in JSON mode, which lets us render replies as metric tiles, charts and tables and attach real actions. Answers flag partial weeks and months on their own and point to the right screen when the data doesn't cover a question.

3. Reasoning with control over cost and speed. Muse Spark reasons before it answers, which shows in the quality of comparisons such as which assistant is growing fastest. The reasoning_effort setting lets us choose: low for the interactive analyst (seconds), more for background checks.

4. One provider for the whole loop. The same model asks the shopper questions in Answer Check, explains the fix in Weekly Fix and powers the analyst, with a single OpenAI-compatible API, so the stack stays small and easy to port, which is how the same features run in both the Shopify app and the web app.

Without the model, Centurion is a set of accurate reports. With it, the merchant asks "what should I fix this week?" and gets the answer, the evidence and a working Apply button in one conversation.

## More from Meta Global AI Developer Hackathon

- [Clipboard History](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5142)
- [NEON](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5143)
- [Proof to Publish](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5144)
- [Alberto Castro](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5146)
- [MediClear](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5147)
- [Ashish Karigar - MemoryOS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5148)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5145. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt