Ollama KNIME - Multimodal Prompting - Car Accident Severity Analysis

This workflow demonstrates how to prompt a local LLM using both text and images.

Textual data from insurance claims and corresponding accident images are imported, combined into row-wise multimodal prompts, and sent to a connected LLM. The model then returns a severity assessment for each accident based on both the text and image inputs.

External resources

• KNIME Forum about multimodal LLM prompts
• Ollama: mistral-small3.1
• Medium: Multimodal Prompting with local LLMs using KNIME and Ollama