Let's say we want to predict insurance frauds. Whenever we obtain an insurance claim, we are provided with a free-form description detailing the loss and a substantial amount of data on the claimant, presented in a tabular format.
How can we utilize both the written loss description and the extensive structured data we have gathered? Should we build two separate models, one for natural language processing and the other for tabular data? Is it possible for Large Language Models (LLMs) to extract insights from tabular data? If so, how? What limitations or pitfalls should we bear in mind?