Im trying to create a bot to search and classify case law. I started by uploading three documents each containing a single case. The case number is in the beginning of each document and in the name of the document. The bot is prompted to know this. However, majority of the time, the bot gives me wrong case number when (for example) giving me a summary of a case. The purpose is to to upload approx 1000 cases so struggling with three is not very encouraging.
Any advice how I could better prompt the bot to understand how to link up a case and case number.
What AI model are you using? Have you tried using GPT 4o and the knowledge exploration feature?
The Knowledge Base chunks documents as uniform chunks (say 250 words each). Then when you ask the bot a question, it pulls the most relevant chunks, which may be parts of a case, but not the entire case. So what is likely happening is that the case number is being orphaned from the rest of the information in the chunk.
You might consider turning the cases into a CSVs file as well with the case numbers as one of the column headers. That would be a helpful supplemental document. Here is a primer video on uploading CSVs into the knowledge base.
1 Like