StitchFlow

Inspiration

I am a knitting and crochet enthusiast, and I often work with patterns written in English, Chinese, and Japanese. Patterns from different countries and designers do not always follow the same conventions. The same abbreviation can mean different things, instructions may mix shorthand with natural language, and it is easy to lose track of where I stopped.

Paper notes did not solve this very well. I wanted one calm place where I could organise a pattern, record exactly how many stitches I had completed, and return days later without having to reconstruct my progress. I also wanted to preserve my own adjustments so that a finished project could become a useful reference for the next one.

That personal problem became StitchFlow.

What it does

StitchFlow is a local-first workflow and progress companion for knitting and crochet projects.

A maker can:

• Create a project manually or explore a prepared sample project
• Organise a pattern into sections, parts, rows, rounds, and steps
• Add, remove, reorder, and edit instructions at any time
• Record progress down to the current completed stitch
• Automatically mark earlier instructions as complete when progress is entered further ahead, while still allowing manual corrections
• View the current position immediately from the project dashboard
• Check terminology across US, UK, and Japanese conventions
• Attach reference images to a project
• Switch between English and Simplified Chinese interfaces
• Adjust text size, contrast, and visual appearance
• Print or save a compact, well-formatted project record as a PDF

The current Version 1.0 deliberately focuses on reliable manual creation and progress tracking instead of pretending that every unstructured PDF pattern can already be parsed accurately.

How I built it

Before Build Week, the repository mainly contained planning documents and an early pattern sample. During OpenAI Build Week, I used Codex with GPT-5.6 Sol to turn that plan into the working product shown in the demo.

I made the product decisions, including the audience, visual direction, information hierarchy, accessibility requirements, terminology boundaries, and progress behaviour. Codex helped me implement and refine the TypeScript domain model, React interface, editing workflow, terminology data, tests, documentation, and deployment.

The application uses React, TypeScript, vinext, Vite, Tailwind CSS, and browser localStorage. It is deployed through OpenAI Sites. Project information stays on the user’s device in Version 1.0, so the live application does not require an account, API key, or paid runtime AI request.

GPT-5.6 Sol was used transparently through Codex as an implementation and reasoning partner. It helped turn product decisions into code, examine edge cases, and accelerate testing and documentation. It is not presented as a hidden runtime API integration.

Challenges I ran into

The hardest challenge was scope.

My original idea included automatic PDF recognition and possibly camera-assisted pattern capture. In practice, knitting and crochet patterns are highly inconsistent. They may contain charts, photographs, custom symbols, regional abbreviations, or instructions written in natural language. Building an unreliable parser in five days would have produced impressive-looking results that makers could not safely trust.

I therefore narrowed Version 1.0 to a workflow that is useful without guessing: structured manual input, editable hierarchy, terminology assistance, and precise progress tracking.

Progress was another deceptively difficult problem. If a maker records five completed stitches in the second round, the first round should usually be considered complete automatically. However, makers do not always follow instructions in order. The model therefore provides helpful automatic completion while keeping every value editable.

I also had to balance a soft frosted-glass visual style with readability for older makers. The interface includes larger text and stronger contrast options instead of treating accessibility as an afterthought.

Accomplishments that I'm proud of

I am proud that StitchFlow became a complete, usable workflow rather than only a visual prototype.

The product now supports the full path from creating a project, structuring its instructions, tracking an exact position, editing the pattern, checking terminology, attaching images, and producing a printable record.

I am especially proud of the current-position model. A maker can open the dashboard and immediately understand where they stopped, without searching through notes or relying on memory.

I also documented the project’s Build Week history, decisions, limitations, tests, and deployment so that the difference between the original plan and the working product is clear.

What I learned

I learned that responsible product design sometimes means reducing the scope.

For craft instructions, a confident but incorrect interpretation can waste hours of work. A transparent manual workflow can therefore be more valuable than an AI feature that guesses silently.

I also learned that progress is not just a percentage. For a maker, “five stitches into round two” is much more useful than “35% complete.” Designing around that real-world mental model made the product substantially clearer.

Working with Codex was most effective when I treated it as an implementation partner rather than asking it to decide what the product should be. Hands-on review, explicit product decisions, and recorded corrections led to a much stronger result.

What's next for StitchFlow

Version 1.0 establishes the project structure, editing system, terminology foundation, accessibility controls, and progress model.

Future versions may explore:

• Assisted PDF import with confidence indicators and mandatory user review
• Cloud synchronisation and optional accounts
• A larger personal terminology library
• Additional interface languages, including Japanese
• Richer export templates
• Sharing and collaboration between makers
• Camera-assisted progress and pattern references

These are roadmap directions rather than features claimed in the current demo. My next priority is to test the existing workflow with more real knitters and crocheters before deciding which automation is genuinely useful.

Built With

• cloudflare
• codex
• next.js
• node.js
• react
• tailwind
• typescript
• vite