Scribby was the project I led at AdventureX 2025: AI that keeps you company while you practise writing, so learning feels as easy as a walk. It was first called 授笔漫步 (Noodling), and later renamed Scribby.
AdventureX was a hackathon in Hangzhou in July 2025, with more than 800 young people under 26 and the theme “Game Everything”. Four of us who love literature formed a team and won third prize in the Zeabur track.
Where the idea came from
I was reading Shen Dan’s book on narratology and stylistics. Narratology splits a novel into two layers: story, what happens, and discourse, how language tells it.
I think everyone has a story they want to tell, but most people struggle to put it into words.
I also remembered a writing class I took as a kid:
Eight of us sat around a table writing essays. Afterwards we passed them around and critiqued each other. Anxiety and self-doubt took up most of my energy, so I couldn’t really focus on getting better at writing.
The problem with learning to write
- People are shy about showing their writing, so they miss out on feedback;
- there’s little practice in revising and polishing, so the writing stays flat.
So my way in was this: getting people to love writing doesn’t have to start with heavy literary theory. It can start with small, fun exercises.
The exercises we designed
One-minute free writing. Write whatever is in your head, without worrying about grammar, structure or what others think. To make people really let go, I designed an odd rule: the delete key doesn’t work, and you get a nudge if you stop for ten seconds. Then you can go again, and the second round asks a guiding question based on what you just wrote.
Writing from a photo. Creative writing classes have an exercise called “describe a photo”. Upload one of your own:
- start from the five senses: what can you smell, what can you hear?
- shift in time: what happened before this photo, and after?
- if you’re stuck, open the photo and the AI circles a detail worth writing about, as a hint;
- I also wanted the AI to play writers like Oscar Wilde or Jane Austen “critiquing” you: underlining what you did well, then rewriting the same scene in their style.
Prompts from your interests. At sign-up you pick a few interests (cats, literature, plants…), answer three random questions, and the AI writes a prompt from them; the next prompt builds on what you wrote and discussed last time.
A story from key frames. Five images, and you write a story with a plot.
All the exercises share one writing interface. In the design doc I imagined it helping people revise: the AI underlines good sentences and offers a little encouragement; on structure, how to move from ABCD to BCAD; on wording, how to pick a more precise verb; and how to bring out one important detail instead of writing a flat list of events.
Later we narrowed the audience to school students and ordered three exercises by difficulty: breaking down a scene (basic), describing a photo (intermediate), and free writing (advanced), plus a library of writing material. The theme colour was a warm orange.
What we built
The hackathon only lasted a few days, so we agreed to build a few small exercises. Tech: Next.js and Tailwind CSS, with Tiptap as the editor; Supabase for the database, sign-in and files; Azure OpenAI’s gpt-4o for feedback and prompts, and Azure AI Vision to recognise people and objects in photos; deployed on Zeabur.
Leading the team
It was my first time leading a team outside school. What I had in mind was:
At least make the team feel they can count on me. I didn’t want everyone to be exhausted, as long as the project was interesting and fun.
One teammate had just graduated. I had him learn HTML, CSS and JavaScript first, then React, “and in the end it was his vibe coding that made our UI stand out.” “I felt like I could be a mentor.” It was also the first time I merged someone else’s PR: “just like when Zhentang merged my code! So satisfying.”
The organisers’ submission system broke, and everyone ended up staying up all night.
Looking back
The result doesn’t matter; the experience does.
On site I also tried an AI design tool, and what stayed with me was that good design needs “its own tone, something recognisable”. I worried that my taste, shaped by office work, would make what I built look standardised.
Scribby was the first prototype of an idea I keep coming back to, a practice space for writing. The idea is still with me.