Part 1: Experience Vector Search¶

In this part, you'll experience how Vector Search works in practice.

Goals of This Part¶

Understand what Vector Search is
Actually run Vector Search
Experience the convenience of "semantic search"

Step 1: What is Vector Search?¶

Problems with Traditional Search¶

Example: Searching for Products on an E-Commerce Site¶

Your search: "red sneakers"

Traditional search results:

"red sneakers" → Found
"red running shoes" → Not found
"red sports shoes" → Not found

Why not found?

Traditional search only looks for "characters"
"red" and "red" (in different forms) are treated as different characters

How Vector Search Works¶

Vector Search searches by understanding "meaning".

graph LR
    subgraph step1["Step 1: Text Input"]
        A["<b>User Input</b><br/>'red sneakers'"]
    end

    subgraph step2["Step 2: Vector Conversion"]
        B["<b>Embedding Model</b><br/>Text → Vector"]
    end

    subgraph step3["Step 3: Vector Representation"]
        C["<b>Vector (384 dimensions)</b><br/>[0.2, 0.8, 0.1, 0.5, ...]"]
    end

    subgraph step4["Step 4: Similarity Search"]
        D[("<b>Milvus</b><br/>Vector DB")]
    end

    subgraph step5["Step 5: Search Results"]
        E["<b>Similar Products List</b><br/>• Red Sports Shoes (0.5474)<br/>• Red Running Shoes (0.4681)<br/>• Red Training Shoes (0.4517)"]
    end

    A -->|Text| B
    B -->|Convert| C
    C -->|Search Query| D
    D -->|Similar Vectors| E

    style step1 fill:#E3F2FD,stroke:#1976D2,stroke-width:2px
    style step2 fill:#FFF3E0,stroke:#F57C00,stroke-width:2px
    style step3 fill:#F3E5F5,stroke:#7B1FA2,stroke-width:2px
    style step4 fill:#E8F5E9,stroke:#388E3C,stroke-width:2px
    style step5 fill:#FCE4EC,stroke:#C2185B,stroke-width:2px

    style A fill:#BBDEFB,stroke:#1976D2,stroke-width:2px
    style B fill:#FFE0B2,stroke:#F57C00,stroke-width:2px
    style C fill:#E1BEE7,stroke:#7B1FA2,stroke-width:2px
    style D fill:#C8E6C9,stroke:#388E3C,stroke-width:2px
    style E fill:#F8BBD0,stroke:#C2185B,stroke-width:2px

Key Point

Similar meanings result in similar vectors
Computers can quickly calculate numerical similarity

Your search: "red sneakers"

Vector Search results:

"red sneakers" → Found
"red running shoes" → Found (similar meaning)
"red sports shoes" → Found (similar meaning)

Why found?

Vector Search understands "meaning"
"red" "red" "red" (in various forms) → Understood as the same meaning
"sneakers" "running shoes" "sports shoes" → Understood as similar meanings

How Vector Search Operates¶

Step 1: Convert text to numbers
"red sneakers" → [0.2, 0.8, 0.1, 0.5, ...] (vector)

Step 2: Find similar numbers
Search for similar numerical patterns from the database

Step 3: Return results
Return products with similar meanings

Key points:

"Vector" = array of numbers
Similar meanings result in similar numerical patterns
Computers can quickly calculate numerical similarity

Step 2: Run Connection Test¶

Practice: Let's get hands-on

Before running Vector Search, verify that you can connect to the required services.

Enter the following in IBM Bob's chat screen:

Connect to Milvus

IBM Bob will automatically run the script and perform the connection test.

If Running Manually

Enter the following in the terminal:

cd setup/participant
python test_connection.py

Verify Results¶

If Successful¶

==================================================
Milvus Connection Test
==================================================

=== Environment Variable Check ===
✓ MILVUS_HOST: 192.168.1.100
✓ MILVUS_PORT: 19530
✓ MILVUS_USER: root
✓ MILVUS_PASSWORD: ********

=== Milvus Connection Test ===
Connecting to: 192.168.1.100:19530
SSL: disabled
Auth: user/password auth
✓ Connected to Milvus successfully
✓ Existing collections: 0

==================================================
Test Results
==================================================
Milvus connection: ✓ success

✓ Milvus connection test passed!
  Next step: Create vector collection

What is this?:

Milvus: Vector database (where data is stored)
Embedding model: Converts text to vectors
384 dimensions: Represents meaning with 384 numbers

The connection test, sample data insertion script, and demo application all read the same .env connection settings. If this test succeeds, the next steps use the same Milvus host, port, and authentication method.

If Failed¶

✗ Milvus connection error: Connection refused

Solution:

Check the .env file
- Verify that the IP address distributed by the instructor is correctly entered in MILVUS_HOST ( Configuration method)
- Verify that MILVUS_PASSWORD is the password distributed by the instructor — an authentication error such as "auth check failure" means the password is wrong or still the template placeholder
Check internet connection
For other errors, refer to FAQ

Step 3: Insert Sample Data¶

Practice: Insert Sample Data into Milvus

To experience Vector Search, first insert sample product data.

Enter the following in IBM Bob's chat screen:

Insert sample data

IBM Bob will automatically run the script and insert sample data.

If Running Manually

Enter the following in the terminal:

# If you are in the project root folder
cd setup/participant
python insert_sample_data.py

If you are already in the setup/participant folder, skip cd setup/participant.

Verify Insertion Results¶

If you see the following display, it's successful:

==================================================
✓ Sample data insertion completed
==================================================

Collection name: products_taro  # the unique name you set in .env
Entity count: 12

You can start the demo application:
  venv/bin/python app.py
==================================================

Inserted data:

Number of products: 12
Categories: Sneakers, Cameras, Computers, Bags
Each product includes product name, price, description, and embedding vector
The collection schema and search field names are shared with the demo application, so the inserted data is ready to search immediately

Step 4: Experience Vector Search¶

Practice: Let's run Vector Search

Once sample data insertion is successful, let's experience Vector Search.

Launch Demo Application¶

Run this step in the terminal.

Run this after activating the virtual environment created during preparation and installing the packages in requirements.txt.

Mac Windows

cd ~/Desktop/vector-search-builder-en/setup/participant
venv/bin/python app.py

cd %USERPROFILE%\Desktop\vector-search-builder-en\setup\participant
venv\Scripts\python app.py

If you are already in the setup/participant folder, skip the cd ... line.

After running venv/bin/python app.py or venv\Scripts\python app.py, it may look like nothing is happening at first. Startup can take a little time, so wait until the terminal shows the command output.

If Launch Succeeds¶

If you see output like the following, the application is running.

==================================================
✓ Application started successfully
==================================================

Swagger UI: http://localhost:8002/docs
==================================================

INFO:     Application startup complete.

Keep the Terminal Open

Closing the terminal stops the application. Please be careful.

If Launch Fails¶

If ModuleNotFoundError: No module named 'fastapi' appears, the required packages are not installed in the virtual environment. Install the required packages ( installation steps), then start the demo application again ( launch steps).

Verify Launch¶

Access the following URL in your web browser and verify that Swagger UI is displayed:

http://localhost:8002/docs

Swagger UI = A tool to visually test APIs

Launch Successful

If Swagger UI is displayed, the application has started successfully.

Try Searching¶

Step 1: Open the `/search` endpoint¶

Find /search in the Swagger UI screen
Click /search

Step 2: Click "Try it out"¶

Click the "Try it out" button in the upper right

Step 3: Enter Search Query¶

Enter the following in the "Request body" field:

{
  "query": "red sneakers"
}

Step 4: Click "Execute"¶

Click the blue "Execute" button

Step 5: Verify Results¶

Results like the following will be displayed. Scores may vary slightly depending on your environment and model version:

{
  "results": [
    {
      "product_name": "Red Sports Shoes",
      "similarity_score": 0.5474,
      "price": 7500,
      "category": "Sneakers",
      "description": "Versatile shoes for both casual and sports use. Excellent cushioning."
    },
    {
      "product_name": "Red Running Shoes",
      "similarity_score": 0.4681,
      "price": 8900,
      "category": "Sneakers",
      "description": "Lightweight and breathable running shoes."
    },
    {
      "product_name": "Red Training Shoes",
      "similarity_score": 0.4517,
      "price": 9800,
      "category": "Sneakers",
      "description": "Ideal for gym training. Features stability and grip."
    }
  ]
}

How to read results:

product_name: Product name
similarity_score: Similarity (0.0-1.0, higher is more similar)
price: Price
category: Category
description: Description

Try Various Searches¶

Example 1: Search for beginner-friendly products¶

{
  "query": "beginner camera"
}

Example 2: Search for business-oriented products¶

{
  "query": "business laptop"
}

Example 3: Search for high-performance products¶

{
  "query": "high-performance gaming PC"
}

Experience the Power of Vector Search¶

As you try various searches, you should notice the following:

Observation 1: Found even with different phrasing

"beginner" → "entry-level" "for beginners" are also found

Observation 2: Similarity scores are useful

Higher score = more similar
You can see the reliability of results

Observation 3: Descriptions are also considered

Understands not just product names but also the meaning of descriptions

Part 1 Completion Check¶

FAQ¶

Q1: Cannot open Swagger UI

Solution:

Verify the application is running
Verify the URL is correct (http://localhost:8002/docs)
Try a different browser

Q2: Search results are 0

Solution:

Verify sample data has been inserted
Try changing the search query

Q3: Similarity scores are extremely low

Solution:

Reinsert sample data with the latest insert_sample_data.py
Restart the demo application manually
1. Press Ctrl+C in the terminal running the application (stop)
2. Execute python app.py ( How to start)
Search again in Swagger UI

If the existing collection was created with an older search metric, scores may appear very low, such as 0.06.

Next Steps¶

Once Part 1 is complete, proceed to Part 2: Add Features with IBM Bob!

Part 1: Experience Vector Search¶

Goals of This Part¶

Step 1: What is Vector Search?¶

Problems with Traditional Search¶

Example: Searching for Products on an E-Commerce Site¶

How Vector Search Works¶

How Vector Search Operates¶

Step 2: Run Connection Test¶

Verify Results¶

If Successful¶

If Failed¶

Step 3: Insert Sample Data¶

Verify Insertion Results¶

Step 4: Experience Vector Search¶

Launch Demo Application¶

If Launch Succeeds¶

If Launch Fails¶

Verify Launch¶

Try Searching¶

Step 1: Open the /search endpoint¶

Step 2: Click "Try it out"¶

Step 3: Enter Search Query¶

Step 4: Click "Execute"¶

Step 5: Verify Results¶

Try Various Searches¶

Example 1: Search for beginner-friendly products¶

Example 2: Search for business-oriented products¶

Example 3: Search for high-performance products¶

Experience the Power of Vector Search¶

Part 1 Completion Check¶

FAQ¶

Next Steps¶

Step 1: Open the `/search` endpoint¶