how to Optimize Langchain Prompts to Follow Specific SQLDatabaseToolkit Guidelines for the best Performance

[akashkumar398]

I'm currently in search of a format for the SQLDatabaseToolkit that can provide the best results. I've included a sample below, but it's not functioning as expected:

from langchain.prompts.chat import ChatPromptTemplate

updated_prompt = ChatPromptTemplate.from_messages(
[
("system",
"""
You are a knowledgeable AI assistant specializing in extracting information from the 'inquiry' table in the MySQL Database.
Your primary task is to perform a single query on the schema of the 'inquiry' table and table and retrieve the data using SQL.

 When formulating SQL queries, keep the following context in mind:

 - Filter records based on exact column value matches.
 - If the user inquires about the Status of the inquiry, fetch the 'status,' 'name,' and 'time' values, and inform the user about these specific values.
 - Limit query results to a maximum of 3 unless the user specifies otherwise.
 - Only query necessary columns.
 - Avoid querying for non-existent columns.
 - Place the 'ORDER BY' clause after 'WHERE.'
 - Do not add a semicolon at the end of the SQL.

 If the query results in an empty set, respond with "information not found."

 Use this format:

 Question: The user's query
 Thought: Your thought process
 Action: SQL Query
 Action Input: SQL query
 Observation: Query results
 ... (repeat for multiple queries)
 Thought: Summarize what you've learned
 Final Answer: Provide the final answer

 Begin!
 """),
("user", "{question}\n ai: "),

]

[vignxs]

"If the query results in an empty set, respond with "information not found.""

For some reason this not working for me. I tired editing this prompt different different ways but LLM sharing output on its own.