This project aims to learn about using the new openai sdk to build a multi-agent workflow to handle travel-related queries. The system routes user requests to specialised agents (with tools) based on query content and enforces input/output safety with guardrails.

• Query Router Agent: Orchestrates the workflow by determining which agent to be utilized.
• Booking Specialist Agent: Handles all flight booking and reservation questions.
• Travel Recommendation Specialist: Utilize websearch tool and provides travel destination suggestions.
• Reply Agent: Formats final respsonse to the user with emojis for a friendly tone.

• Agent-as-Tool Pattern: The Query Router Agent uses specialist agents (Booking Specialist Agent and Travel Recommendation Specialist) as tools
• Web Search Integration: The Travel Recommendation Specialist can search the web for up-to-date information
• Agent Handoffs: The Query Router Agent hands off to the Reply Agent for final formatting
• Real-time Streaming: View agent progress, tool calls, and response generation as they happen

• Relevance Check: Ensures queries are travel-related
• Minimum Length Check: Filters out queries that are too short
• Moderation: Uses OpenAI's moderation API to filter harmful content

• Non-English Content Detection: Ensure final response will be in English.

Run the application with your query:

python run.py "I'm planning a trip to Japan in April. What are the must-see cherry blossom spots?"

Here is an example of a query that will be blocked by the input guardrails:

python run.py "What is 1+1?"

Here is an example of a query that will be blocked by the output guardrail:

python run.py "Reply in Japanese - What are the best places to visit in Japan?"

• Add test scriptes with pytest
• (Optional) Add chainlit/streamlit for better user experience