Organize chatbot outputs with memory

Designing Agentic Systems with LangChain

Dilini K. Sumanapala, PhD

Founder & AI Engineer, Genverv, Ltd.

Streaming multiple tool outputs

    Full graph workflow.

 

  • Print outputs with multiple tools
      - Agent: [Tool 1: answer]
  - Agent: [Tool 2: answer]
  • Print user query and enable memory
      - User: [Query]
  - Agent: [Tool 1: answer]

  - User: [Follow-up query]
  - Agent: [Tool 1: follow-up answer]
Designing Agentic Systems with LangChain

Streaming multiple tool outputs

from langchain_core.messages import AIMessage, HumanMessage


config = {"configurable": {"thread_id": "1"}}


# Create input message with the user's query
def multi_tool_output(query):

    inputs = {"messages": [HumanMessage(content=query)]}


    # Stream messages and metadata from the chatbot application
    for msg, metadata in app.stream(inputs, config, stream_mode="messages"):


        # Check if the message has content and is not from a human
        if msg.content and not isinstance(msg, HumanMessage):
            print(msg.content, end="", flush=True)

    print("\n")
Designing Agentic Systems with LangChain

Test with multiple tools

Check dynamic tool assignment
multi_tool_output("Is `Stella won no wallets` a palindrome?")

multi_tool_output("What happened on April 12th, 1955?")

The phrase or word 'Stella won no wallets' is a palindrome.

Yes, the phrase "Stella won no wallets" is a palindrome.


April 12th, 1955, is notable for several reasons, particularly in the context of science and 
technology. On this date, the first successful polio vaccine developed by Dr. Jonas Salk 
was announced to the public. This vaccine was a significant breakthrough in the fight against 
poliomyelitis, a disease that had caused widespread fear and numerous outbreaks, particularly 
affecting children. The announcement marked a turning point in public health and led to 
widespread vaccination campaigns that ultimately contributed to the near-eradication of polio...
Designing Agentic Systems with LangChain

Follow-up questions with multiple tools

# Print the user query first for every interaction
def user_agent_multiturn(queries):
    for query in queries:
        print(f"User: {query}")


        # Stream through messages corresponding to queries, excluding metadata 
        print("Agent: " + "".join(msg.content for msg, metadata in app.stream(
                {"messages": [HumanMessage(content=query)]}, config, stream_mode="messages")


            # Filter out the human messages to print agent messages
            if msg.content and not isinstance(msg, HumanMessage)) + "\n")


queries = ["What happened on the 12 April 1961?", "What about 10 December 1948?", 
"Is `Mr. Owl ate my metal worm?` a palindrome?", "What about 'palladium stadium?'"]
user_agent_multiturn(queries)
Designing Agentic Systems with LangChain

Full conversation output

  • Historical events
User: What happened on the 12 April 1961?
Agent: On April 12, 1961, Yuri Gagarin, a Soviet cosmonaut, became the first human to travel 
into space aboard the Vostok 1 spacecraft...

User: What about 22 November 1963?
Agent: December 10, 1948... marks the Universal Declaration of Human Rights (UDHR).
  • Palindrome check
User: Is `Mr. Owl ate my metal worm?` a palindrome?
Agent: The phrase or word 'Mr. Owl ate my metal worm?' is a palindrome...

User: What about 'palladium stadium'?
Agent: No, the phrase `palladium stadium` is not a palindrome...
Designing Agentic Systems with LangChain

Let's practice!

Designing Agentic Systems with LangChain

Preparing Video For Download...