System prompts

🤖 Building Specialized Chatbots (AWS Support Specialist)

Goal: To transform a general chat interface into a specialized chatbot that adheres to strict content and behavioral rules.

Required Chatbot Behavior (Do's & Don'ts):

• Must Do:
• Directly answer the user's question.
• List various methods for hosting services (e. g. , Postgres on AWS).
• Guide the user through initial setup steps.
• Mention related AWS services.
• Must Not Do:
• Mention competitor solutions.
• Answer questions unrelated to cloud services (e. g. , personal advice, recipes).

Implementation Strategies:

• Option 1: User Message Requirements (Inefficient)
• Requires listing all "do's and don'ts" within the user's input.
• Drawback: Extremely tedious and difficult to maintain for every possible query.
• Option 2: System Prompt (Recommended)
• Provides explicit guidance on how the model should respond.
• Achieved by assigning a specific role to the model (e. g. , "You are an AWS cloud support specialist").
• This forces the model to adopt the persona's constraints (e. g. , refusing non-cloud topics and avoiding competitors).

Technical Implementation Details:

• System Prompt Function: The prompt is passed into the model's function call (e. g. , the converse function).
• Constraint: The system prompt must contain at least one character.
• Best Practice (Refactoring): The chat function should be refactored to accept the system prompt as a parameter, rather than hardcoding it, to allow for flexible use.

Takeaways

• Specialized chatbots must adhere to strict behavioral rules, such as answering specific questions and avoiding off-topic or competitor content.
• The System Prompt is the recommended implementation strategy, providing explicit guidance on how the model should respond and defining its persona (e. g. , "AWS cloud support specialist").
• The System Prompt forces the model to adopt constraints, ensuring it only addresses relevant topics and follows defined operational guidelines.
• Best practice requires refactoring the chat function to accept the System Prompt as a parameter, allowing for flexible and maintainable use instead of hardcoding it.