TimeXL introduces a novel framework that combines time series data with contextual information through an LLM-in-the-loop approach, delivering both improved accuracy and interpretable results.

• Multi-modal integration - Leverages prototype-based encoders with three collaborating LLMs to process both time series and auxiliary data
• Enhanced predictive power - Produces more accurate forecasts by incorporating contextual signals often overlooked by traditional methods
• Explainable results - Provides transparent rationale for predictions, enabling more informed decision-making
• Engineering applications - Offers valuable capabilities for monitoring complex systems, predictive maintenance, and process optimization

Explainable Multi-modal Time Series Prediction with LLM-in-the-Loop