LiftBuilder - cs428TAs/w2025 GitHub Wiki

Overview

Working out can be difficult, especially when you don't have a planned workout routine. A lot of people hitting the gym just move about without a real plan, but this strategy doesn't produce long-term gains.

Enter LiftBuilder, an application that powers decision-making at the gym. Using queries to LLMs like Claude and OpenAI, LiftBuilder takes advantage of the volumes of workout-related data to generate a workout to fit each user's needs. It carefully curates workout plans based on the needs of each user.

Team

  • Talmage Bird
  • Nikolas Earl
  • Rebekah Erikson
  • Sam Gwilliam
  • Austin Warnick

Deliverables

  1. Organization Chart and Role Requirements
  2. Project Requirements
  3. Pert Chart and Gantt Chart
  4. Architecture and Design
  5. SQA/Test Plan

Status Reports

Pitch

Background

Working out is difficult these days, especially when you don't have a workout routine planned out for you. I work out as much as I can, and I'll usually just jump around and work on whatever I want to work on, but this strategy hasn't netted me many gains or results. There is a breadth of knowledge out there about what kind of exercises and movements will net the best results, but a lot of that knowledge is gatekept or scattered by the services that charge you subscription fees. I want to use open source software to help me easily generate a workout that meets my needs for any given day.

Idea

Luckily, large language models like Claude and ChatGPT have access to large amounts of workout-related data. This data can be pulled out through careful queries to these models, and it can be compiled into a user interface that is descriptive and friendly. Enter LiftBuilder, a single-page application that uses LLM APIs to power decision making for a workout matrix given a list of supported workout movements and types.

Features

Planned

  • User Profiles with standard username/password authentication
  • Profiles store workout preferences, workout history, and data to quantify what works and what doesn't
  • Interface to show a workout plan and its inherent details

Possible

  • An interactive workout interface to keep track of lift status (like in the Hevy app)
  • A basic prompt interface with a pre-seeded LLM chat geared towards fitness and wellness
  • Stronger authentication with sso like Google and Facebook or with passkeys

Architecture

Frontend

Planned
  • React, NodeJS, Static Site with Vite packaging
  • Hosted in GitHub Pages for now?
Possible
  • Hosting in AWS, Azure, or some other cloud platform
  • Flutter, React Native, or some other native device app framework
  • PHP and all the stuff that comes with it?

Backend

Planned
  • Hosted on AWS in a containerized Fargate module
  • Redis LLM query cache with MySQL RDS for user storage
  • Rest API to connect back and front-end
Possible
  • Azure or other cloud platform
  • Another cache db of some kind to manage llm queries
  • Neo4j graph persistence to generate user profiles

Proposed by Nikolas Earl

⚠️ **GitHub.com Fallback** ⚠️