Sockrates AI Designer

Title: AI Sockboard Designer: Custom Sock Concepts from a Logo in One Minute

Summary I built an AI design tool for a custom sock manufacturer. It turns a client's logo into a presentation-ready "Sockboard" PDF with five woven-sock design concepts and two label options. The team used to build these by hand in Illustrator. Now it takes about a minute and roughly $0.01 per board.

What I built

Logo-to-board pipeline: upload any logo (PDF, AI, EPS, SVG, PNG or JPG) and get a finished board as a vector PDF.

Manufacturing rules built in: the client's knitting rules (5-colour limit, 2-colour cuffs, matching toe and heel, no logos on toe or heel) are stored once and enforced automatically on every design.

Smart logo handling: picture files are redrawn as vectors, background boxes are removed, and small text like taglines is left off the socks so logos stay legible when knitted.

Design variety: a library of 40 preset styles (argyle, houndstooth, tartan, rugby stripes and more) rotates, so boards don't repeat.

Instructions followed: notes like "use navy" or "include tractors" are applied, then checked by a second AI pass.

Simple interface for non-technical staff: guided three-step setup, live progress, one-click "redesign this sock", saved versions, and instant edits to colour names and Pantone codes.

How it works The AI makes the design decisions but never draws. It writes a structured design plan, which is checked against the manufacturing rules and then drawn by code into an exact vector PDF. That keeps logos and brand colours precise, which image-generation models can't guarantee.

Tech: Next.js, TypeScript, Vercel AI Gateway (Gemini / Claude), MuPDF, Ghostscript WASM, headless Chromium, Zod

Outcome: the client's team now produces first-draft sock proposals in minutes instead of hours, which makes outbound prospecting at scale practical.