Analysis of data from a wearable using natural language.

Support:

• Garmin (In-progress)
• TODO

Usage:

> source .env
> PYTHONPATH=./ python app/main.py --task "Plot my sleep times and steps daily for last week in the same plot"

An executed ipynb notebook will be created in the artifacts directory with the desired task completed. Examples of the output are in the examples directory.

Dev guide

ENV

• Set all required env vars in .env
• source .env && python app/main.py

Langfuse

• Clone and run server: https://langfuse.com/self-hosting/local
• Set up API keys via UI

TODOs

• observability with text
• handling long text outputs in jupyter cells
• handling errors in executing notebook
• partial execution of notebook to avoid repeated login
• shorter cell outputs
• verify output loop
• observability with images
• saving traces
• profiling speed of solver
• dataset of tasks for evaluation
• handle case where login fails