# About Vivian

## What Vivian is

Vivian is a document accessibility platform. It analyzes PDF libraries against WCAG 2.1 Level AA and the PDF/UA-1 structural checkpoints, then remediates them: structure tags, reading order, alt text, table structure, form labels, document title, and language. Every run ends in an audit-ready report that lists each issue found, what was done about it, how confident the system was, and what still needs a person.

## Who it is for

Universities, state and local governments, healthcare organizations that receive federal funding, and enterprises facing ADA Title II, Section 504, and Section 508 obligations with far more PDFs than any team could fix by hand. Manual remediation takes two to four hours per document; a large institution holds tens of thousands of them.

## How it works

Vivian parses each PDF with its geometry intact, infers headings, reading order, tables, links, and language, and maps every failure to its WCAG success criterion. AI proposes alt text with a confidence score. Fixes at or above your threshold (0.85 by default) are applied and logged; anything below it, plus medical imaging and technical diagrams regardless of score, goes to a human reviewer who approves, edits, or rejects it. The result is validated against PDF/UA-1 before the report is written.

## What we hold ourselves to

Automated analysis is not legal certification, and every report says so. Automation rates are measured per document class by a reproducible benchmark rather than asserted. Customer documents are never used to train models and are deleted automatically when your retention window ends.

## Who runs it

Vivian is an early-stage, independent product built by Karthik Subramanian in Berkeley, California, and runs on Cloudflare infrastructure. To talk about a pilot or anything else, see the [contact page](https://tryvivian.pages.dev/contact).