LLM powered Workflow based Time Series Analysis

This repository contains the code for this (blog post)[https://medium.com/@benedikt_heidrich/can-we-do-time-series-analysis-with-llm-powered-workflows-using-sktime-12b19cf39376].

To run the code you need:

• to add an Mistral API key to a .env file in the root of the repository like this:

MISTRAL_API_KEY = <your_api_key>

• to install the dependencies using uv.

Then you can run the notebooks:

• workflow_with_sktime.ipynb, which contains three runs of the workflow:
  • Forecasting of load airline data
  • Forecasting of electricity demand of New South Wales
  • Anomaly detection on a very simple synthetic time series.
• consistency_analysis.ipynb, contains simple code to analyse how consistent the LLM is in selecting estimator.

The agentic workflow is implemented in src/workflow/agent.py.