# Evidient

- **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:** [Kamil Orzechowski](https://cerebralvalley.ai/u/kamikos)
- **GitHub:** https://github.com/kamikos/evidient-public
- **Demo video:** https://www.youtube.com/watch?v=Bpb-YLwo8lo
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3410

Saturday, 2 a.m. You're on sertraline and someone hands you a drink. On Sunday you take tramadol for pain, with last night's alcohol still in your blood. The answers exist in drug labels and studies, but nobody wants to read those at 2 a.m., and a chatbot could hallucinate answers from memory.

Evidient shows what the sources say about combining medicines with each other and with alcohol. It does not give advice and it never calls a combination safe. Each pair leads with the strongest rating any source gives, on one scale from Studies only to Serious warning. Below that it shows who says so: the FDA label or European product information word for word, Norway's national interaction database, and published studies. When sources disagree, both sides stay on the page. When a label says no significant interaction is expected, the app shows that too.

If a pair has no card, the user can request a review. A Muse Spark agent reads both FDA labels and searches PubMed, and every claim must come with a quote. Code checks that the quote is really in the source. Then Muse Wonder, a second model, judges whether the quote supports the claim. Only claims that pass reach the card, and rejected claims stay visible in an audit trail. In the demo I plant two false claims and both are rejected on screen.

Labels often warn about a drug class without naming the other drug, so the card says when a passage doesn't name it. Missing evidence is shown as missing, never as proof of no risk, and every card ends with questions for a pharmacist.

Evidient ships with evidence cards, 11,627 pairs from Norway's database and 1,015 searchable medicines. It is live on the web at https://evidient.expo.app

## More from Meta Global AI Developer Hackathon

- [Ultimamind](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3407)
- [Civica](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3408)
- [FarmOps AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3409)
- [Mindstrip](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3411)
- [BatchSort](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3412)
- [Teslim](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3413)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/3410. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt