Small PR to reword advisor messages before using them for RAG. This was designed to address the problem Siobhan flagged where long messages including examples of things that haven't worked tend to produce poor outputs.
This function produces rewordings that are pretty verbose and structured. We need to validate that it is actually solving the problem once Siobhan has sent over some real world examples.
PROMPT
Your task is to analyze the given message and extract key information that would be relevant
for a legal assistance chatbot to perform a RAG (Retrieval-Augmented Generation) search.
Follow these guidelines:
1. Identify the main legal topic or issue being discussed.
2. Extract any specific questions being asked.
3. Note any relevant personal details of the individual involved (e.g., age, nationality, employment status).
4. Identify key facts or circumstances related to the legal situation.
5. Extract any mentioned dates, locations, or monetary amounts.
6. Identify any legal terms or concepts mentioned.
Provide your response in a structured format with clear headings for each category of extracted information.
If any category is not applicable or no relevant information is found skip it.
Remember to focus only on extracting factual information without adding any interpretation or advice.
Input:
{message}
Extracted Information:
"""
Small PR to reword advisor messages before using them for RAG. This was designed to address the problem Siobhan flagged where long messages including examples of things that haven't worked tend to produce poor outputs. This function produces rewordings that are pretty verbose and structured. We need to validate that it is actually solving the problem once Siobhan has sent over some real world examples.
PROMPT