def get_judge_engine() -> QueryPipeline:
    # Setup LLMs
    judge_llm = Settings.llm
    worker_llms = {}
    num_workers = int(os.getenv("NUM_WORKERS", "5"))
    for i in range(num_workers):
        worker_llms[str(i)] = Settings.llm
    
    # Construct the query pipeline
    p = QueryPipeline(verbose=True)

    # Define the pipeline nodes.
    module_dict = {
        **worker_llms,
        "worker_query": WORKER_PROMPT_TMPL,
        "judge_query": JUDGE_PROMPT_TMPL,
        "llm_judge": judge_llm,
        "join": ArgPackComponent(),
    }
    p.add_modules(module_dict)

    # Add links between nodes 
    for i in range(num_workers):
        p.add_link("worker_query", str(i))
        p.add_link(str(i), "join", dest_key=str(i))

    p.add_link("join", "judge_query", dest_key="context_str")
    p.add_link("judge_query", "llm_judge")
   
    # Generate visualization
    net = Network(directed=True)
    net.from_nx(p.dag)
    net.save_graph("rag_dag.html")

    # Return the final pipeline.
    return p

def useless_tool() -> int:
    """This is a uselss tool."""
    return "This is a uselss output."

useless_tool = FunctionTool.from_defaults(fn=useless_tool)