video1.0<iframe src="https://www.loom.com/embed/1b5db9494ed948e19683caa7dd746e99" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/1b5db9494ed948e19683caa7dd746e99-2f86bc5cee6cab33.gif301.967This Loom explains the architecture and workflow for an AI tool that audits Figma designs and prepares them for smoother code generation. The user inputs a Figma URL, the tool retrieves node details in JSON via the Figma API, and eight audits validate items like naming, SVG layers, spacing, and contrast ratio to produce a violation report cached in memory. It then launches a vision-agent API using a React loop that runs up to five times with the Jina 2.5 Flashcore vision model, while users supply company guidelines in markdown for the agent to search and rank relevant chunks. In a demo, the tool reports mostly contrast ratio and layout auto layout issues, inspects specific frame and text nodes, and provides hierarchy and typography guidance such as differentiating competing primary buttons.