Ascend

• 

  Main screen

• 

  Solo Mode

• 

  Graph for topic - recursion

• 

  First Question

• 

  Failing a node leads to wrecked screen showing reason for failure and what to review

• 

  Clearing all nodes unlocks boss node

• 

  Final completed screen with time elapsed

• 

  Battle mode

• 

  Host and participant

• 

  Battle start for 2 participants

• 

  Final winner displayed for all participants

• 

  Correct answer explanation on failure for one user

Inspiration

I have always enjoyed learning new concepts, and gamifying the experience of learning helped me stay motivated long-term. Whether it's using Pomodoro timers, checking off to-do list tasks and rewarding myself with an episode of my favorite drama or anime, I found that turning progress into a game makes learning more enjoyable. Combining this with my love for the game skribbl, I was inspired to create a graph-based knowledge game where you can play solo or even battle it out with friends, to level up and test your knowledge.

What it does

Ascend is a study tool shaped like a roguelike. Type in any topic and GPT-5.6 generates a live prerequisite skill tree for it: a graph of the concepts you actually need to understand, in order, ending in a "boss" concept that represents real mastery. You answer questions to clear each node. Nodes stay locked until their prerequisites are cleared, so you can't skip ahead! If you get something wrong, Ascend traces back through the graph to find the actual concept that you find shaky, even though that's not the question you just failed. Most study tools tell you what you got wrong, Ascend tries to tell you why, structurally. You can also play it in battle mode. Share a room link and race a friend through the same generated tree in battle mode - same questions, live leaderboard, first to the boss wins.

How I built it

I built it using GPT-5.6 (via OpenAI structured outputs) for graph and question generation and Codex for scaffolding, feature implementation, and debugging real integration issues along the way. I used React along witht tailwindCSS for the frontend.

Challenges I ran into

• Designing a prerequisite graph that makes sense for different subjects
• Figuring out how to render the graph before realizing that I could use dagre and ReactFlow
• Debugging issues encountered while playing - like displaying a meaningful traceback for the shaky node (tweaking the prompt), and several other UI issues

Accomplishments that I'm proud of

• Building an AI-generated knowledge graph instead of relying on static course content
• Successfully integrating GPT into the core gameplay instead of using it as a chatbot
• Making learning feel fun like a game
• Delivering this project within a week

What I learned

• Prompt engineering has a huge effect on the consistency and quality of AI-generated questions.
• User experience matters as much as model quality

What's next for Ascend

• Smarter difficulty adjustment based on long-term performance.
• A profile with levels that increase based on the difficulty of the topic
• Team battles and tournaments
• Mobile app

Built With

• fastapi
• openai
• react
• tailwindcss