AccessAI

Your 24/7 Autonomous Accessibility Auditor.

Comment

Inspiration

The internet is broken for 1.3 billion people. 96% of the top 1 million homepages fail basic accessibility standards. Developers don't fix these issues because auditing is tedious and invisible until a lawsuit happens. We realized current tools just complain at developers. We built AccessAI to fix the problem autonomously browsing like a visually impaired user and patching the code before a human even notices.

What it does

Live Automation: Our agent (powered by TinyFish) autonomously browses the website, streaming real-time logs and capturing screenshots of accessibility failures. Intelligent Triage: The system analyzes the raw automation logs and visuals to identify valid violations, filtering out noise and structuring them into clear engineering tickets. Owner Reconnaissance: Yutori's Research API simultaneously scours the web to identify the technical owners (names/emails) responsible for the site. Instant Handoff: With one click, findings are exported to Linear. The system automatically maps severity to Linear priorities (e.g., "High" → "Urgent") and sets tickets to "Active" on your sprint board.

How we built it

TinyFish (Automation Engine): TinyFish visit each site and run accessibility checks and performs different audit frameworks Yutori (All Four APIs) Discovery (Yutori Scouting): Identify new startups and add them to the audit queue. Visual Validation (Yutori n1): Using the n1 pixels-to-actions model, we test if UI elements are navigable via keyboard. We also use this to generate structured JSON for the linear tickets Enrichment (Yutori Research): Simultaneously, the Research API identifies the site’s tech stack and technical owner, ensuring the final Linear ticket is assigned to the right person with stack-specific instructions. Deep Navigation (Yutori Browsing): It evaluates accessibility visually as opposed to TinyFish that uses the DOM Linear (Workflow Integration): We built a direct integration that maps our structured findings into Linear Issues. The system automatically sets priorities (Urgent/High) and moves tickets to "Active," creating an instant workflow for product teams. Cline (The Builder): We used Cline as our autonomous pair programmer to write the complex glue code connecting the TinyFish SSE streams to our backend pipelines. Cline handled the error handling and type definitions for our API routes. Macroscope (The Scribe): We integrated Macroscope to automatically generate the context and summary for our Pull Requests, saving developers from writing documentation.

Challenges we ran into

Hallucinations: Early versions tried to fix unbroken code. Yutori did not return the html.

Accomplishments that we're proud of

Smooth collaboration among three teammates, handling different agent tasks and UI

What we learned

Workflow over detection. Define idea before generating code. Collaboration matters

What's next for AccessAI

Jira & Asana Support: expanding beyond Linear to support enterprise workflows. Auto-Remediation: allowing the agent to not just ticket the issue, but propose the actual code fix in a PR.

Built With

+ 1 more
Share this project:

Updates