Part 3: Verification and Review¶

In this part, you'll verify that the features added in Part 2 work correctly together and request a code review from IBM Bob.

Goals of This Part¶

Test the added features
Request a code review from IBM Bob
Verify code quality

Step 1: Test the Added Features¶

Practice: Let's test the added features

Test the features added in Part 2 without repeating the same verification point.

If testing was already completed in Part 2

During Part 2, IBM Bob may automatically create test scripts and complete the tests after you click the run button. Even in that case, use this step to double-check that the added features work as expected.

Test viewpoints

In Swagger UI, one search response can show all added response fields. Therefore, this section separates testing into response field verification and price filter behavior verification.

Test 1: Overall Response Field Verification¶

Procedure¶

Open Swagger UI (http://localhost:8002/docs)
Open the /search endpoint
Click "Try it out"
Enter the following:
```
{
  "query": "red sneakers"
}
```
Click "Execute"

Verify results:

{
  "results": [
    {
      "product_name": "Red Sports Shoes",
      "image_url": "https://example.com/images/red-shoes.jpg",
      "similarity_score": 0.5474,
      "price": 7500,
      "category": "Sneakers",
      "description": "Versatile shoes for both casual and sports use. Excellent cushioning.",
      "recommendation_reason": "Related to the search content (similarity: 0.5474)"
    }
  ]
}

Verification Points¶

image_url field exists
URL is in correct format (starts with https://)
recommendation_reason field exists
Recommendation reason is displayed in clear language
Existing product_name, price, category, and description fields are still displayed

Procedure¶

Open /search in Swagger UI
Click "Try it out"

Enter the following:

{
  "query": "sneakers",
  "min_price": 5000,
  "max_price": 10000
}

Click "Execute"
Verify results: All product prices are within the 5000-10000 yen range

Verification Points¶

Only products within the specified price range are displayed
Products outside the range are not displayed
image_url and recommendation_reason are still displayed

Optional: Try Various Price Ranges¶

// High price range
{
  "query": "camera",
  "min_price": 50000,
  "max_price": 100000
}

// Low price range
{
  "query": "camera",
  "min_price": 0,
  "max_price": 20000
}

Test Completion Check¶

Verified image_url and recommendation_reason in one search response
Verified that the price filter excludes products outside the range
Verified that the added features work correctly together

Step 2: Request a Code Review from IBM Bob¶

Practice: Request a code review

Have IBM Bob check the code quality.

What is Code Review?¶

Code review = Verifying code quality

Purpose:

Find bugs
Discover improvement points
Learn best practices

Request Review¶

About the /review command

IBM Bob has a /review command. However, IBM Bob 1.0.3 does not support passing a file name to target a specific file, such as /review <file>. To review a file, ask in natural language as shown below, such as "Review app.py".

Enter the following in the chat input field and press Enter:

Review app.py

IBM Bob's Code Review

When you request a review, IBM Bob will execute a code review.

Start with app.py because it is the API entry point. If Feature 1 changed the product data or Milvus schema, also review the related shared files:

Review schema.py
Review insert_sample_data.py

Review IBM Bob's Review Results¶

IBM Bob will perform analysis like the following:

app.py Code Review¶

✅ Good Points¶

1. Structure and Architecture¶

Clear RESTful API design using FastAPI
Type-safe request and response models with Pydantic
Appropriate separation of concerns through imports from common.py and schema.py

2. Multilingual Support¶

Japanese / English switching with the msg() function
All user-facing messages support multiple languages

3. Error Handling¶

Appropriate HTTP status codes (503, 500)
Exception handling with try-except blocks
User-friendly error messages
Modern startup/shutdown handling with a lifespan handler
Input validation for query and top_k with Pydantic Field constraints

4. Feature Completeness¶

✅ Vector search
✅ Price filter (min_price, max_price)
✅ Automatic recommendation reason generation
✅ Health check endpoint

5. CORS Settings¶

Allows access from the frontend

⚠️ Improvement Suggestions¶

1. Use of global variables (lines 24-25)¶

# Current
embedding_model: Optional[SentenceTransformer] = None
collection: Optional[Collection] = None

# Recommended: use dependency injection or application state
app.state.embedding_model = None
app.state.collection = None

2. Add validation for the price range added in Part 2¶

from pydantic import BaseModel, Field, model_validator

class SearchRequest(BaseModel):
    query: str = Field(min_length=1)
    top_k: int = Field(default=5, ge=1, le=100)
    min_price: Optional[int] = None
    max_price: Optional[int] = None

    # Recommended addition
    @model_validator(mode="after")
    def validate_price_range(self):
        if self.min_price is not None and self.max_price is not None:
            if self.max_price < self.min_price:
                raise ValueError("max_price must be greater than or equal to min_price")
        return self

3. Improve logging¶

import logging

logger = logging.getLogger(__name__)

# Use logger instead of print()
logger.info(f"Connecting to Milvus: {host}:{port}")
logger.error(f"Failed to connect: {e}")

4. Strengthen security¶

# Restrict CORS settings for production
app.add_middleware(
    CORSMiddleware,
    allow_origins=["http://localhost:3000"],  # Specific origins only
    allow_credentials=True,
    allow_methods=["GET", "POST"],  # Required methods only
    allow_headers=["*"],
)

5. Optimize performance¶

# Add caching for frequently searched queries
from functools import lru_cache

@lru_cache(maxsize=100)
def get_query_embedding(query: str):
    return embedding_model.encode([query], normalize_embeddings=True)[0].tolist()

6. Add tests¶

# tests/test_app.py
from fastapi.testclient import TestClient

def test_search_endpoint():
    client = TestClient(app)
    response = client.post("/search", json={"query": "red sneakers"})
    assert response.status_code == 200
    assert "results" in response.json()

📊 Code Quality Score¶

Item	Score	Comment
Readability	9/10	Clear naming and appropriate comments
Maintainability	8/10	Modularized, but uses global variables
Extensibility	9/10	Easy to add new features
Error handling	8/10	Basic handling is implemented
Security	6/10	CORS settings are permissive
Performance	7/10	No caching

Overall rating: 8.0/10 - Recommended improvements before moving to production

🎯 High-Priority Improvements¶

Medium: Add validation for the price range added in Part 2
Medium: Implement logging for debugging and monitoring
Low: Tighten CORS settings for production
Low: Replace global variables with application state

The current code is fully functional and works well as a demo application!

Review Improvement Points¶

Review the improvement points IBM Bob pointed out.

Common improvement points:

Adding log output
Strengthening error handling
Adding documentation comments
Creating test code

How to Treat Improvement Suggestions

IBM Bob's review results may include improvement suggestions intended for production environments. Global variables, CORS settings, caching, and additional tests are items to consider when moving toward production.

Request Improvements (Optional)¶

If there are points you want to improve, request IBM Bob:

To add logging:

Add log output to the search processing in app.py

To add validation:

Add min_price and max_price validation to SearchRequest in app.py

To review CORS settings:

Restrict the CORS settings in app.py for production

To add tests:

Add tests for the search endpoint in app.py

Part 3 Completion Check¶

Tested the features added in Part 2 using non-overlapping verification points
Requested a code review from IBM Bob
Reviewed schema.py and data insertion files too if the product data schema changed
Reviewed review results and understood improvement points

FAQ¶

Q1: Test fails

Solution:

Verify application is running
Verify changes are saved
Restart the application manually
1. Press Ctrl+C in the terminal running the application (stop)
2. Execute python app.py ( How to start)

Q2: Review results are not displayed

Solution:

Verify the review request was entered correctly
Verify file name is correct
Restart IBM Bob

Step 3: Environment Cleanup¶

Practice: Clean up the virtual environment

Deactivate the virtual environment used in the hands-on and delete the hands-on folder.

Purpose of Cleanup¶

In this hands-on, you worked inside the vector-search-builder-en folder on your desktop. When you finish working, first deactivate the virtual environment. Then delete the vector-search-builder-en folder from your desktop to remove the hands-on files, project-local venv, and configuration files together.

Request IBM Bob to Clean Up¶

Enter the following in IBM Bob's chat input field:

Deactivate the Python virtual environment and delete the vector-search-builder-en folder from the desktop.

How cleanup works

deactivate only exits the virtual environment in the current terminal. Installed packages remain inside the venv folder.

deactivate is not included in folder deletion. It resets the current terminal before you delete files.

Run deactivate first, then delete the vector-search-builder-en folder created on your desktop. This also removes the project-local venv and configuration files.

If cleaning up manually

Execute the following in the terminal:

Mac Windows

deactivate
cd ~/Desktop
rm -rf vector-search-builder-en

deactivate
cd %USERPROFILE%\Desktop
rmdir /s /q vector-search-builder-en

If deleting only venv

Usually, deleting the entire vector-search-builder-en folder is enough. If you want to keep the project files and remove only the virtual environment, execute the following:

Mac Windows

deactivate
cd ~/Desktop/vector-search-builder-en/setup/participant
rm -rf venv

deactivate
cd %USERPROFILE%\Desktop\vector-search-builder-en\setup\participant
rmdir /s /q venv

Q: What if I want to keep the venv?

If you plan to use the technologies learned in this hands-on in the future, you don't need to delete the venv folder. Just run deactivate and reuse it later.

Cleanup Completion Check¶

Deactivated the virtual environment
Deleted the vector-search-builder-en folder

Next Steps¶

Once Part 3 is complete, proceed to Summary!

Part 3: Verification and Review¶

Goals of This Part¶

Step 1: Test the Added Features¶

Test 1: Overall Response Field Verification¶

Procedure¶

Verification Points¶

Test 2: Price Filter Behavior Verification¶

Procedure¶

Verification Points¶

Optional: Try Various Price Ranges¶

Test Completion Check¶

Step 2: Request a Code Review from IBM Bob¶

What is Code Review?¶

Request Review¶

Review IBM Bob's Review Results¶

app.py Code Review¶

✅ Good Points¶

1. Structure and Architecture¶

2. Multilingual Support¶

3. Error Handling¶

4. Feature Completeness¶

5. CORS Settings¶

⚠️ Improvement Suggestions¶

1. Use of global variables (lines 24-25)¶

2. Add validation for the price range added in Part 2¶

3. Improve logging¶

4. Strengthen security¶

5. Optimize performance¶

6. Add tests¶

📊 Code Quality Score¶

🎯 High-Priority Improvements¶

Review Improvement Points¶

Request Improvements (Optional)¶

Part 3 Completion Check¶

FAQ¶

Step 3: Environment Cleanup¶

Purpose of Cleanup¶

Request IBM Bob to Clean Up¶

Cleanup Completion Check¶

Next Steps¶