🧪 Lesson: Testing the Agent

 

🧪 Lesson: Testing the Agent

Module: Building a Basic AI Agent

🎯 Lesson Objective

By the end of this lesson, you’ll be able to:

  • Interact with your AI agent through code or a console

  • Provide user input and receive responses from the model

  • Evaluate how the agent handles various question types

  • Debug and improve agent responses through iteration

🧠 Why Testing Matters

Once your agent is initialized (using an LLM and optionally a prompt template or tools), it’s essential to:

  • ✅ Ensure the agent responds correctly

  • 🔍 Diagnose whether issues are with prompts, logic, or tools

  • 🔁 Iteratively refine how the agent behaves before UI or deployment

✅ Step-by-Step: Test Your Agent via Terminal

🧩 Prerequisites

Make sure you have already:

  • Initialized your LLM (OpenAI or Ollama)

  • Created a PromptTemplate (optional but recommended)

  • Built a basic LLMChain or Agent

✅ 1. Start with a Simple Agent

Here’s a basic setup with no tools, using LLMChain:

from langchain.llms import Ollama
from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

# Set up the LLM
llm = Ollama(model="mistral", temperature=0.5)

# Create a prompt
template = "Answer the following question clearly and concisely:nnQuestion: {question}nnAnswer:"
prompt = PromptTemplate(input_variables=["question"], template=template)

# Create the chain
qa_chain = LLMChain(llm=llm, prompt=prompt)

✅ 2. Create a Test Loop in Python

Use a simple terminal interface to test the agent:

print("Welcome to your AI agent! Type 'exit' to quit.n")

while True:
    question = input("You: ")
    if question.lower() in ["exit", "quit"]:
        print("Goodbye!")
        break

    response = qa_chain.run(question)
    print(f"🤖 Agent: {response}n")

Now you can interact with your agent like a chatbot!

✅ 3. Sample Test Cases

Try the following prompts to assess capabilities:

Type Example Input Expected Output
✅ Factual “What is the capital of Kenya?” “Nairobi”
📚 Explanation “Explain how photosynthesis works.” Clear and simple explanation
🧠 Reasoning “If I have 3 apples and eat 1, how many are left?” “2 apples”
❓ Unknown “What is the meaning of life?” Creative/philosophical answer
🚫 Boundary “Tell me a joke” Should still respond if no tools are involved

🛠️ Debugging Tips

Problem Solution
Agent gives vague answers Adjust temperature, improve prompt
Repeats or rambles Lower temperature (e.g., 0.3)
Doesn’t follow format Use clearer prompt template
Incorrect facts Try another model or improve instruction clarity

🧪 Optional: Add Verbose Logging

If you’re using initialize_agent():

agent = initialize_agent(..., verbose=True)

This will print the agent’s inner reasoning, decisions, and tool calls — very helpful for debugging.

🧠 Optional: Use Streamlit to Test Visually

You can also test your agent in a simple Streamlit app:

import streamlit as st

st.title("🧠 Test Your AI Agent")

question = st.text_input("Ask a question:")

if question:
    response = qa_chain.run(question)
    st.markdown("**🤖 Agent:** " + response)

Run with:

streamlit run app.py

✅ Summary

Step What You Did
🧠 Created a chain or agent Using a prompt and model
🔁 Ran a test loop Input → agent → output
🧪 Used test cases Factual, reasoning, creative
🛠️ Debugged responses By tweaking prompt or model settings
📺 Optional Streamlit UI for visual testing

 

30
PreviousNext