← ALL PROJECTS
UC Berkeley AI Hackathon, 2023 · Scale AI sponsored track

Enable

A tool that turns a plain language description of a task into a ready-to-print prosthetic design, aimed at cutting the cost and complexity out of getting assistive technology.

My RoleGenerative model, math + LLM integration
Team3 people
Built WithGPT, MATLAB, Next.js, 3D printing, fine-tuning
Try ittinyurl.com/sorocalhacks

Overview

Prosthetic devices are typically custom built per person, which makes them expensive and slow to get. Enable takes a natural language prompt, something like "I want to play the guitar," and turns it into an STL file: a 3D-printable prosthetic design ready to go, with no engineer or designer required in the loop.

What I worked on

  • Generative model. Built the core generative pipeline linking the language model's output to the mathematical model that produces the physical design.
  • LLM to geometry. Helped turn natural language requirements into parameters and, eventually, into physical dimensions for the printed part.

How it works

Each generated prosthetic is modular and attaches to a single central mount powered pneumatically. The team fine-tuned a model on synthetic training data, since real-world prosthetic design data is scarce, and used that model to map a plain-language prompt through to a printable STL file. The frontend was built in Next.js.