ai-workout-trainer

Awesome resource to get started with free API access https://github.com/cheahjs/free-llm-api-resources

Active todo:

Get V0.1 of the app working in streamlit with chat
- formatted output properly with editable table
Install Docker locally to build app image
Get pathing working properly in app script
Update the prompt for proper formatting
- Add a finish for workouts :]
Add format to st.dataframe.
- Make it editable
Connect Linux machine to NAS
- Setup and connect to postgres DB
Self-host langfuse

Consider wrapping everything in Langfuse
- https://langfuse.com/
We need chat history to update our workout list before we get started
- Take in previous workout and modify with the given user's update
- How do we handle this with the client?
  - If first query, do generate()
  - else, do update_response()
  - Have a counter to keep track of this?
  - Or can we just throw it all in the chat history and have the LLM figure it out, probably tbh
Front end to display generated workouts
Prompt to give upper body, lower, core, etc. These can came from buttons in the app
Motivational quote to accompany workout
Prompt formatting for correct output
- Agnetic framework to find videos and explain moves from youtube
Gather knowledge base of workouts
Host on NAS
Built in rep counter
Create mobile app
Save / export / import workouts from previous sessions
- Save to local backend database on NAS
Have backlog of workouts to call to
- Inject into prompt to help standardize workouts
- This needs clarified. It all depends on how well the LLM performs without existing workouts to work off
- The LLM might do well enough from just the knowledge base and itself
High temperature for variability of workouts
As a side note, this could also apply to BJJ
- I want to drill guard passing today, what should I work on
- Could use metadata for inside or outside passing