I'm building a software application. I want to use a LLM as the basis. I will finetune the model with about 20,000 pages of legal text. Specifically laws all around the country. I will then use the trained model to answer help companies create compliant products and services. At the moment I am unsure as to the best way to go about it. My initial thought was to use gpt 3.5 and then further train it with the 20,000 pages of text. The text will be broken down by state and federal agencies. This text will be added to as new laws and regulations are passed.

​

The goal is for the model to only answer based on the data set it's finetuned with. The data will be broken down by state. For example if they want information about loan requirements, they will ask the system and it will return with an outline of the requirements for loans by state. It will respond in a way that's easy to understand.

​

My thinking is once I have all the data collected, using Langchain to fine tune the LLM. Am I on the right path here?