SwitchboardAI

SwitchboardAI is an intelligent intent routing system developed based on LangGraph and Ollama. It acts like a highly efficient switchboard, automatically identifying whether a user's input is an "emotional need" or a "logical inquiry," and accurately connecting the conversation to the most appropriate agent for processing.

Live Demo

demo.mp4

🌟 Core Features

1. Intelligent Intent Routing: Utilizes Pydantic structured output to accurately determine the user's intent.

2. Dual Agent Architecture:

• Emotional Agent: Focuses on emotional support and empathetic responses, suitable for psychological counseling or emotional expression.

• Logical Agent: Focuses on fact analysis and information retrieval, with web search capabilities.

3. Multi-engine Dynamic Search: Integrates with the Tavily and DuckDuckGo search tools, automatically triggering when the Logic Agent needs external information.

4. State Cycle Management: Uses LangGraph to maintain session state, supporting information aggregation and reprocessing after tool calls.

🏗️ System Architecture

The system workflow is shown in the diagram below:

1. Classifier: Receives input and classifies it as emotional or logical.

2. Emotional Path: Guides to a healing dialogue; ends upon completion.

3. Logical Path: Guides to factual answers. If a search is detected, a Tool Node is invoked to retrieve the results, which are then summarized by the logical agent.