How Wojciech Jaszczak Built an AI-Powered Children's Book Platform in a Weekend

The MVP took two days. By the time this interview was published, childbook.ai had generated over 2,500 books and was growing at roughly 10% week-on-week — with no marketing spend.

Wojciech Jaszczak, a Senior Fullstack Engineer at X-Team, built childbook.ai alongside his wife — a self-taught frontend developer — after the two entered an AI hackathon together. The premise was simple: give the platform a short description of a story and its characters, and the AI returns a personalized story with illustrations and voice-overs in under five minutes. The execution was anything but simple.

In this story, Jaszczak explains how a hackathon weekend turned into an ongoing product, which technologies made a two-day MVP possible and what the platform's roadmap looks like from here.

From Hackathon Prompt to Working Product

The idea came out of a practical observation. Jaszczak and his wife knew that AI could generate both text and images. Personalized children's stories seemed like a natural combination of those two capabilities — useful, fun and technically interesting enough to build in a weekend.

They had prior experience with Stable Diffusion from an earlier project and knew it could generate high-quality images if prompted correctly. They also knew GPT-3 could generate coherent text. "We combined both to see how far we could get," Jaszczak says. The MVP was fully functional at the end of the weekend.

The hardest part of the build wasn't getting the text to work — it was the illustrations. "AI is very unpredictable and prompts that give very good results for some stories can give very bad results for others," he says. The team had to experiment extensively with DALL-E and Stable Diffusion parameters, and with text generation, to arrive at something consistent and visually appealing. Quality has improved significantly since launch, and the upgrade from GPT-3 to GPT-4 marked a turning point: "It actually returns what you'd expect now."

The stack that made the two-day build possible: Next.js, Prisma, DALL-E for illustrations, Stable Diffusion to post-process them, GPT-3 for text generation, tRPC for typed end-to-end API calls, Redis with bullMQ for job queuing, PostgreSQL for story storage, AWS S3 for images, Stripe for payments, and Tailwind CSS with React Headless UI for components. Using TypeScript across the full stack kept bugs low and iteration fast. "It was a very pleasant experience that I'd recommend to anyone starting a new project," he says.

Maintaining Quality at Scale

With 2,500 books generated and a growing subscriber base, image consistency has become the defining engineering challenge. The platform generates base images with DALL-E (SDXL) and then post-processes them twice to align styling and visuals. The double pass exists because entering too much detail into a single prompt causes the image to lose detail — so the team generates the image multiple times rather than trying to get everything right in one shot.

Books are currently capped at 12 pages. That limit is intentional: it's the threshold at which Jaszczak and his wife can reliably maintain quality. Expanding beyond it without sacrificing consistency is one of the near-term engineering problems they're working through.

The platform currently has around 150 daily users from organic traffic alone. That number is expected to grow once a set of larger features — currently in development — are finished and a proper marketing effort begins.

What the Roadmap Looks Like

Three features are in active development. The first is a DreamBooth implementation that would allow subscribers to upload photos of friends or family members, train the AI on those images and then generate a story with those people illustrated inside it. "A perfect gift for any occasion," Jaszczak says.

The second is physical book printing. The platform already generates illustrations at 4K resolution — the quality needed for print — so once the formatting work is done, subscribers will be able to order physical copies of their stories.

The third is expanding the page count beyond 12 while keeping quality consistent across the full book.

Beyond those near-term milestones, Jaszczak sees several longer-range directions for the product. One is becoming synonymous with personalized children's stories; another is expanding into other genres, like fantasy. A third possibility appeals to him for a different reason: a more curated mode where AI functions as a creative copilot rather than the main driver, "a way to release the creativity of people who may not have the artistic skills, but who have the imagination to create something beautiful."

Ready to build work you're proud of? Apply for an open role at X-Team.