AI Transparency

Flaaze deploys generative AI in a product advisor. This page says plainly how, in the spirit of the voluntary transparency practices encouraged by California AB 2013.

1. The models

Language models and embeddings come from the providers listed on the providers page. The engine around them is owned by Kavod AI LLC and licensed exclusively to Flaaze.

2. What the system generates

The system generates short answers in natural language. Each answer sits on one level of the Response Ladder. The lowest levels use client authored store answers and what a validated certificate of analysis shows, with no model generation for the store answers. Higher levels describe interaction checks, category education and experience fit, and, only where graded human evidence supports it, a cited recommendation.

3. What it does not generate

It does not diagnose, prescribe, or say that a product treats, cures or prevents a disease. It does not state that something is safe because no data was found. Where nothing is documented it says no data on file.

4. Fixed refusals

When a question calls for a prohibited or unsupported claim, the system declines with a fixed explanation that does not change from answer to answer, and, where permitted, offers general education. A refusal is a feature, not a fault.

5. Citation grading

Where a statement rests on literature, the citation is shown with a grade of the strength of the evidence. Public literature comes from PubMed. A statement without adequate evidence is not made.

6. Human review of rule packs

Rule packs, the fixed refusal copy and the classification of restricted products are written and reviewed by people. Compliance review is required for restricted products before go live.

7. Limits

AI answers can be wrong or incomplete. Every answer carries an AI disclosure and is not medical advice.

8. Reporting a problem

If an answer looks wrong or unsafe, write to hello@flaaze.com with the page it appeared on and what was said. We review reports and correct the rules, content or evidence involved.