Because compliance usually finds you first.

FDA enforcement is, by law, public. Every warning letter, recall, and enforcement action is published through the openFDA API and mirrored on FDA.gov. The information a brand needs to understand its regulatory exposure is sitting in the open — it's just scattered across thousands of records, in formats built for analysts, not founders.

Meanwhile, most brands don't learn they have a violation until a letter arrives. By then the failure mode is fixed in place: a disease claim on a product page, a structure-function statement with no disclaimer, a marketing line that was never compliant from the day it shipped. The gap isn't legality — it's visibility.

We built ComplianceGuard to close that gap. We pull the public docket and turn it into a per-brand risk map: your public pages on one side, what the FDA has already punished on the other. Instead of a wall of raw records, you get a short list of the specific claims on your own site that look like the ones regulators have acted on.

It's a small, distributed team of regulatory-data engineers and copy specialists. We're not regulators, and we're not lawyers. We're a reading layer on top of public data — built to give brands a fair, fast look at where they actually stand.

The pipeline runs in four stages. Each one is built to keep the output grounded in real, verifiable information rather than generalization:

1. Fetch public pages. We retrieve the pages you submit — homepage, product pages, about page — and read only content that is already publicly available.
2. Extract claims. A multi-model AI pipeline with specialized roles for classification, copy generation, and reasoning reads each page and extracts health claims, structure-function statements, and disclaimer coverage.
3. Cross-reference the docket. Every extracted finding is compared against our index of 2,039 openFDA enforcement records in food, drug, and device categories, refreshed from api.fda.gov on 2026-10-17.
4. Produce the report. You receive a risk-scored inventory (1–10 per finding) with the supporting evidence quote and a compliant rewrite where one is possible.