Behind the story packs: branching design, writing decisions, and lessons from building interactive classics.
Why I hesitated after Day 32, and the product/UX decisions we made on February 18, 2026 to move forward with more clarity.
Three consecutive sessions focused on fixing image drift, restructuring storypacks, and redeploying Jekyll with a complete pipeline.
Shipped The Wizard of Oz to the website and set up a lightweight Mem0 memory workflow to keep decisions searchable across sessions.
Analyzed story quality issues across existing packs, then generated all art assets for The Wizard of Oz. Learned why choices need dilemmas, not preferences.
Designed and built a collectible card reveal experience for story endings. Flip animations, share functionality, and bilingual support included.
Found multiple issues in I, Robot and fixed them fast. Turned ad-hoc fixes into a cleaner workflow with validation and regeneration.
Generated all personality cards, resolved image loading, and shipped the full I, Robot story pack to the website.
Finished I, Robot's visual assets and discovered a critical bug: the story generator wasn't creating personality card data. Fixed the workflow and saved future stories from the same fate.
Asking AI to generate a complete branching story in one shot? Too ambitious. Breaking it into two phases—linear story first, then branch expansion—solved everything. Sometimes constraints unlock creativity.
Built a collectible personality card system using AI image generation. The cards look stunning—authentic Pokemon/MTG aesthetics. But AI struggles with text. Now solving for both beauty and accuracy.
Completed The Great Gatsby interactive story with 49 AI-generated scenes, fixed character consistency issues, and built a system to prevent recurring deployment mistakes. Sometimes the best lessons come from fixing what broke repeatedly.
Implementing the story quality fixes revealed an unexpected optimization: 75% reduction in content while improving user experience. Sometimes fixing one problem solves three.
20 days of building: 3 story packs, bilingual support, 16 days ahead of schedule. Then user testing revealed a critical flaw that changes everything.
Debugging Astro script injection and managing session state for a cleaner user experience.
Why I spent a day organizing files and writing a social card generator.
Sometimes the biggest usability issues are invisible to you until someone points them out. Today: fixing code blocks, language consistency, and the most important button I almost forgot.
Hit Claude Pro's 200k token limit repeatedly, ending sessions mid-task. Built a complete token management system with automated log maintenance, emergency protocols, and optimization strategies that doubled work capacity.
Deployed Pride and Prejudice and Frankenstein. Fixed the MBTI personality calculator to show descriptions on ending pages. Plus, I started using Google's Gemini 3 Pro alongside Claude.
Discovered critical illustration quality issues with Pride and Prejudice story pack. Here's how I fixed multiple twins, giant characters, and improved AI image generation through iterative prompt engineering.
Found a critical flaw in AI-generated stories: only 3 endings instead of 12+. The MBTI personality feature couldn't work properly. Here's how I improved the story generator.
My CLAUDE.md file was too big and slowing down AI responses. Reorganized project documentation to reduce token usage by 95%.
Korean translation was done, but the desktop layout was broken and images weren't showing up. Spent the day fixing responsive design issues.
Finally got the Jekyll & Hyde interactive story working with MBTI personality tracking and multiple endings.
Setting up GitHub repository, connecting to Vercel, and deploying What If Classics to whatifclassics.com for the first time.
Building the website to test What If Classics stories in a real browser. Setting up Astro for static site generation and running the first local test.
Adding illustrations to interactive stories using Gemini's image generation API with reference images for consistency.
Developing the story pack generator and ensuring quality through visualization for What If Classics.
From concept to architecture: Designing the story pack generator and service architecture for What If Classics.
Domain registration and social media setup: Establishing What If Classics' online presence.
Finding the perfect name with AI partner: Domain brainstorming and brand naming
Exploring revenue potential and market scale with AI partner: Validating the business model for What If Classics.
Planning the project with AI partner: Using Claude Code to build faster and smarter as a solo founder.
From a newsletter email to an interactive story platform: How I'm using Claude Code to build What If Classics as a solo founder.
Exploring why classic novels continue to captivate readers in our fast-paced, technology-driven world.