Inspiration

We noticed a growing gap: while AI tools like Lovable allow anyone to generate websites through natural language prompts, there are no tools designed for kids—or students with learning differences—to meaningfully engage with them. Platforms like Scratch and Code.org empowered children to create visually and interactively, but nothing exists yet for this new generation of AI-first web design. Lovabuddy was born out of a desire to create a bridge: a fun, visual, and accessible way for any child to turn their imagination into a working AI-built website.

What it does

Lovabuddy is a Chrome extension and on-page AI co-designer for kids. Children can draw their website ideas or describe them in plain language. Our assistant then asks multiple choice questions—based on their input and drawing—to gather more details in a fun, game-like way. From there, we automatically generate a high-quality prompt and build the website, with no need to touch any external AI tool. Every experience is guided by a cheerful, animated mascot and narrated using expressive TTS, making it inclusive for kids of all backgrounds and learning styles.

How we built it

We used a combination of React for our frontend overlay and Chrome extension APIs to inject context-aware functionality directly into webpages. Content scripts inspect live DOM elements, allowing the system to provide page-aware editing suggestions and intelligent feedback. A background agent orchestrates multiple LLM-powered modules (including drawing interpretation, prompt refinement, and layout correction) while our overlay renders structured responses in real time. We emphasize accessibility with adjustable contrast, reduced motion, and voice narration, and the child’s input is always respected—whether it comes from a sketch, a sentence, or a simple tap on an answer card.

Challenges we ran into

Creating a seamless pipeline from child-friendly input (drawings, plain language) to precise AI prompting was challenging. We had to balance the number and type of questions asked so that kids stay engaged while still giving us enough information to produce quality websites. Getting emotive text-to-speech from Gemini's 2.5 model via Google Cloud would sometimes cause a 5-second delay before the audio would start. We solved this by cacheing and synchronously processing the audio while questions are being generated by our agent. Finally, making our UI both accessible and fun, especially when making considerations for students with special needs.

Accomplishments that we're proud of

We’re proud to have created a truly end-to-end system that allows children to go from sketch to fully generated AI website without needing to touch any external tool. Its ease of use means that anyone can use it and have fun. We messed around with it a lot, doing random prompts, and it was really cool to see the creativity how our random ideas were interpreted.

What we learned

We learned how to distill technical processes like prompt engineering into playful, educational modules that children can intuitively understand and enjoy. We also learned to develop multiple specialized agents working in unison to manage context handoffs, and maintain coherence.

What's next for Lovabuddy

We plan to expand Lovabuddy into a full learning platform. In the future, kids will be able to unlock guided tutorials. We'll also introduce classroom and caregiver modes for teachers and parents to track progress and curate safe outputs. As AI-native creation becomes the norm, we want Lovabuddy to be the companion that grows with the next generation.

Also v0buddy!?!?!?