Route Genie

Inspiration

As avid climbers and engineers, we noticed that setting routes on kilter boards is slow, inconsistent, and requires expert knowledge. Climbing gyms often struggle to generate varied and challenging routes for each of our skill sets. We thought: “What if we could harness the power of AI to automate this process, making climbing more accessible, fun, and perfect for our skill level?”

This inspired us to combine AI, robotics, and user interfaces to create a system that can generate climbing routes from a simple prompt and deploy them to a kilter board in real-time.

What We Learned

Working on this project taught us a lot about:

Prompt engineering: Translating human-readable route requests into structured AI prompts.
AI integration: Using the Gemini API and then our own model to generate route JSON data.
Embedded systems & Bluetooth: Converting AI output into commands the kilter board can interpret.
User interface design: Building a responsive UI to preview and send climbing routes.
Problem-solving: Handling mismatched data formats, timing issues, and real-world constraints like hold placement.

We also got a deeper appreciation for how AI can bridge the gap between physical and digital experiences.

How We Built It

The system has four main components:

Prompt Parsing & Engineering

Users input a natural-language description of the route (difficulty, style, start/finish).
We preprocess the prompt to optimize it for the Gemini API, ensuring consistent and high-quality outputs.

AI Route Generation (Gemini API -> our own model)

The engineered prompt is sent to the API.
The API returns a JSON describing the holds, including position, color, and type.
Example JSON snippet:

   {
     "id": "1133",
     "color": "cyan"
   }

Conversion to Bluetooth Commands

The JSON is converted into Bluetooth packets compatible with the kilter board.
Each hold’s coordinates and color are encoded into a compact format for transmission.

UI & Deployment

The UI previews the route on-screen, showing hold positions and colors.
Users can click “Send Bluetooth” to deploy the route to the kilter board.
Real-time feedback ensures the AI route matches expectations before climbing.

Mathematically, hold positions are scaled from the AI output using:

$$ cx_{board} = cx_{json} \times w_{board} + x_{offset}, \quad cy_{board} = cy_{json} \times h_{board} + y_{offset} $$

where $w_{board}$ and $h_{board}$ are the physical board dimensions.

Challenges

Data alignment: Ensuring JSON holds coordinates matched the physical board layout.
Bluetooth reliability: Handling intermittent connections and ensuring correct packet delivery.
Prompt quality: Engineering prompts to consistently produce valid climbing routes were nontrivial.
User experience: Designing a UI that was both intuitive and accurately reflected AI output.

Despite these challenges, we iterated quickly, tested often, and ended up with a fully functional system that can generate and deploy routes with a single click.

Reflection

This project showed us the power of AI in physical applications. By connecting AI output to real-world hardware, we created a tool that could revolutionize climbing gyms. Beyond the technical skills, we also improved our teamwork, problem-solving, and pitching abilities.