- OSS4AI June 2025 Hackathon Repo for team
Gym#
Team Name
- Gym# (gym-sharp)
Team Members
- Aaryaneil Nimbalkar
- Darrell Ross
- Austin Motley
- Michael William Wijaya
AI Enhanced Workout Form Training
- Leverage AI with multi-modal input to interpret images of people working out
- Detect what workout form they are doing
- Make recommendations on improved form
- Allow video input
Slides available for download here
- Gradio - quickly build a demo API
- Python Imaging Library - used for opening, editing, and converting images in python
- Torch - tensor formatting
- LangChain/LangGraph - form recommendations
- Upload images
- Record single frame by asccesing webcam
- Paste image in
- LangChain - react agent
- LangGraph - workflow
- Results - display recommendation
We ran the model locally and the overall image recognition had a lot of errors - not recognizing different forms.
This repo contains a Makefile and Dockerfile for building out the container that runs the project. Two different containers were designed.
The first container tried to use N8N for workflow building while also providing JupyterLab for experimentation. While it does build, this proved to be too much work.
Build
make build
Run
make run
Browse
- N8N: http://localhost:5678
- JupyterLab: http://locahost:8888
- App: http://localhost:7860
The second smaller container just runs the app while mounting local files so that we can experiment more easily.
Build
make docker-build-small
Run
make docker-run-small
Browse