Kidpofy
A platform that turns children's handwriting into real digital fonts. Behind the playful interface lies a sophisticated computer vision pipeline, AI agents for quality control and font assembly, RAG-powered typography knowledge, and Claude-driven guidance that helps parents capture the best possible characters.
Visit kidpofy.com(OTF, TTF, WOFF)
Turning messy handwriting into pixel-perfect typography.
Children's handwriting is beautifully imperfect — and that's exactly what makes it worth preserving. But converting those wobbly, inconsistent letterforms into a functional digital font is a deeply complex technical problem. Professional typographers charge hundreds of euros and take weeks because every character needs: contour extraction from noisy input, stroke smoothing without losing the child's personal style, baseline and sizing normalization across 80+ glyphs, proper kerning pairs, and assembly into formats that work on every device.
The real challenge was building intelligence into the pipeline. Children draw characters inconsistently — some too small, some rotated, some barely recognizable. The system needed to handle all of this gracefully: guiding the child during capture, assessing quality in real-time, and knowing when a character needs to be redrawn versus when it's "perfectly imperfect."
And it all had to feel like a game, not a test. The user experience needed to be joyful enough that a 6-year-old would enjoy the process.
Four agents powering the pipeline.
Each stage of the font creation process is owned by a specialized agent. Together they form an autonomous pipeline that turns raw strokes into production-quality typefaces.
Capture Guide Agent
Powered by Claude, this agent guides the drawing experience in real-time. It analyzes each character as the child draws it, detecting common issues — character too small, drawn outside the baseline zone, strokes too thin to reproduce. Instead of rejecting input, it provides encouraging, child-friendly feedback: "That's a great A! Can you try making it a little bigger?" The agent uses Claude's structured outputs to classify character quality and decide whether to accept, suggest a retry, or offer specific guidance.
Vision Processing Agent
The core computer vision pipeline. Takes raw character captures and runs them through a multi-stage process: contour detection and noise removal, stroke vectorization using Bézier curve fitting, weight normalization to ensure consistent line thickness, baseline alignment using a RAG-retrieved knowledge base of typographic rules for each character class. The agent knows that a lowercase 'g' has a descender while a 'b' has an ascender — and adjusts positioning accordingly.
Font Assembly Agent
Takes processed vectors and assembles them into a complete typeface. This agent handles kerning pair calculation (using a RAG-retrieved database of common letter combinations and spacing rules), glyph metrics, font metadata, and multi-format export. It generates OTF for desktop use, TTF for broad compatibility, and WOFF for web embedding. Each output is validated against font specification standards before delivery.
Quality & Delivery Agent
The final gate. This agent renders test strings using the generated font and uses Claude's vision capabilities to assess overall quality: Are characters consistently sized? Does the font feel cohesive? Are any glyphs broken or malformed? It generates a preview image, handles secure font file delivery, and triggers the postcard and merchandise integrations for physical products using the child's custom typeface.
Typography knowledge, embedded.
The system encodes decades of typographic expertise into a RAG knowledge base that agents query at every stage of the pipeline.
A vector-indexed knowledge base of character anatomy: ascenders, descenders, x-heights, stroke widths, and typical proportions for each glyph. Agents retrieve this context to make processing decisions specific to each character class.
RAG-retrieved spacing rules for 500+ common letter combinations. The Font Assembly Agent queries this database to calculate optical spacing that looks right to the human eye, not just mathematically even.
Every completed font is rendered and evaluated by Claude's vision API. The model assesses cohesion, readability, and character consistency — flagging specific glyphs that may need regeneration.
Claude generates encouraging, age-appropriate feedback during the drawing process. It balances quality requirements with keeping the experience fun — never making the child feel like they made a mistake.
Built with Claude Code.
Claude Code accelerated every phase of development — from prototyping the vision pipeline to deploying the production system.
Used Claude Code to rapidly iterate on the computer vision pipeline — testing contour detection algorithms, Bézier curve fitting approaches, and normalization strategies in parallel.
A project-specific slash command that generates test fonts from sample input, renders preview strings, and validates output against font specification standards — all in one command.
Project rules ensure consistent vector output formats across agents, enforce typographic terminology in code and comments, and mandate quality thresholds for every pipeline stage.
The full system.
Backend & Infrastructure
AI & Vision
Font Engineering
Need a computer vision
pipeline?
We build systems that process visual data into structured, production-ready outputs. Let's talk about your use case.