What if your text and images could be analyzed by the same LLM entirely offline? 
@mlauber71 shows how to use KNIME + Ollama to enable multimodal prompting with local models like mistral-small 3.1 and analyze car insurance claims and images — secure, private, and powerful. 
Enjoy the data story!
PS: 
#HELPLINE . Want to discuss your article? Need help structuring your story? Make a date with the editors of Low Code for Data Science via Calendly → Calendly - Blog Writer
There seems to be a discrepancy between the post and the actual model output for one of the accidents.
@rfeigel good catch 
will correct this. Not sure an actual insurance CEO will use this model. This was to demonstrate that the use of text and images in a local LLM is possible even on a local machine. Actually implementing them in production requires some effort and in might involve using a major LLM secured against data leaks and hosted in a region where you have legal protection.
Very nice project. I think its a good demo of multimodal capabilities. A production model would have to handle geographic variability of labor rates. To do some crude testing, you could add a locale to the accident description. I live in the US. In my case you could compare New Jersey (high cost) with Vermont (relatively lower cost.) Such a comparison would do some testing of the LLM’s “blackboxiness”.
