SearchWorker

Introduction

Search Workers retrieve real-time data from the web or databases, and generate coherent and user-specific responses. The searched results enable them to address complex and time-sensitive queries effectively by integrating up-to-date information with their foundational knowledge.

Implementation

With a similar structrue to the RAGWorker, instead of retrieving from local documents, it retrieves from the internet instead through Tavily's search engine API.

Tavily API Integration

You can get an API key by visiting this site and creating an account.

Class Attributes

description: "Answer the user's questions based on search engine results"
llm: LLM model to be used for path generation
action_graph: LangGraph StateGraph

Visualization

Instantiation

On instantiation, the LLM model and the StateGraph is created with the nodes and edges declaration.

def __init__(self):
        super().__init__()
        self.action_graph = self._create_action_graph()
        self.llm = ChatOpenAI(model=MODEL["model_type_or_path"], timeout=30000)

def _create_action_graph(self):
    workflow = StateGraph(MessageState)
    # Add nodes for each worker
    search_engine = SearchEngine()
    workflow.add_node("search_engine", search_engine.search)
    workflow.add_node("tool_generator", ToolGenerator.context_generate)
    # Add edges
    workflow.add_edge(START, "search_engine")
    workflow.add_edge("search_engine", "tool_generator")
    return workflow

Execution

Search Retrieval

The search_engine node calls the SearchEngine.retrieve method which first converts the chat history dependent query to a independent query and runs it via the Tavily's search engine API. It returns the series of URL and its content as the context.

Generation

The generation part is the exact same as RAGWorker's generation.

Sample Conversation

searchworker_function_sample.json

Bot: Hello! How can I help you today?
You: I want to know the latest gaming result of Formula 1.
Bot: The latest gaming result from Formula 1 is that Charles Leclerc won the United States Grand Prix, with Ferrari achieving a one-two finish. Max Verstappen finished behind the Ferrari drivers.

You can checkout the LangSmith Trace for the execution process. The Search Worker first searches online resources relevant to the user's query and then generates the final response.