TripChat revolutionizes the process of travel planning by implementing a hierarchical agent structure. In this system, agents with specialized roles collaborate to curate a comprehensive and personalized trip plan. Each agent focuses on different aspects of travel, such as discovering local events, booking hotels, finding the best flight options, and more. This organized division of labor enhances the efficiency and effectiveness of the trip planning process.
- Hierarchical Agent Structure: Different teams and agents work together for a unified goal, enhancing the efficiency of travel planning.
- Microsoft’s Autogen Framework: Utilizes the Autogen framework for multi-agent systems and GroupChat for agent communication.
- Langchain Framework: Low-level agent implementation for direct interaction with external tools and APIs.
- Shared JSON Object: A collaborative approach where each agent updates a shared object, akin to a collective Google Doc.
- Diverse Agent Types: Includes proxy, assistant, and GptAssistants, each with specific roles and capabilities.
- Install Dependencies: Start by installing necessary dependencies from
requirements.txt
. - Configure API Keys: Modify the
keys.json
file with the relevant API keys. We use this Booking.com API. - Running the Application: Navigate to the
src/
folder and runpython3 main.py
.
After running main.py
, provide a query detailing your trip plan requirements (dates, destinations, etc.). The system will guide you through the process.
- GroupChat Dynamics: Agents communicate in rounds, with a manager coordinating the discussion.
- Role-Based Communication: Agents are aware of their team members' capabilities and communication is structured accordingly.
- Speaker Selection Logic: A novel approach to select speakers, enhancing the hierarchy and communication flow.
- Functionality: Agents use specific functions for updating the shared plan and interacting with external tools.
Special thanks to CS194 Berkeley course staff for their continued support.
TripChat - Streamlining travel planning through innovative agent collaboration.