We built ClimbBright because we wanted a “coach in your pocket” for climbing, something that can turn a wall photo into real training guidance, especially when you’re projecting alone or short on time.
We built a two-stage computer vision pipeline where it detects holds and classifies hold types. The web app overlays those holds on the wall image and generates suggested sequences plus a difficulty estimate and coaching notes (Gemini). We wrapped it with a full-stack experience (FastAPI for inference, Node/Express + Mongo for login/sessions and saving uploads).
The biggest challenges were making all the services run smoothly on one machine, keeping the frontend and backend data format consistent, and avoiding environment issues, like the wrong Python interpreter missing dependencies. Along the way, we learned that a correct starting point, such as reliable setup, clean scripts, stable APIs, and fallbacks, matters just as much as the model itself.
Log in or sign up for Devpost to join the conversation.