Mock LLM Reference Guide

This guide explains how the Mock LLM works in the sample project and provides a reference for available products and matching rules.

Overview

The Mock LLM is a simplified implementation designed for quick testing and demonstration purposes. It simulates AI agent behavior by extracting keywords from user requests and matching them against available products in the in-memory shopping repository.

How Mock LLM Works

Configuration

The Mock LLM behavior is configured in sample-agent/src/main/resources/application.yml under the agent.mock.strategies section:

agent:
  mock:
    enabled: false  # Enable when using --profile mock-llm
    strategies:
      - name: shopping-search
        intent: "search products"
        keywords: [ "search", "find", "look for", "show" ]
        tool-server: "shopping"
        tool-name: "search_products"
        param-rules:
          - param: "keywords"
            source: "user_input"
            pattern: "search\\s+(.+)"
        response-template: "Found the following products: {result}"
        error-template: "Search failed: {error}"

Matching Algorithm

The Mock LLM uses a strategy-based matching algorithm:

1. Keyword Matching: Each strategy has a list of keywords. The user input is split into words and matched against these keywords.
2. Strategy Selection: If any keyword from a strategy matches the user input, that strategy is selected.
3. Parameter Extraction: The selected strategy uses regex patterns to extract parameters from the user input.
4. Tool Invocation: The extracted parameters are passed to the configured tool server and tool name.
5. Response Formatting: The result is formatted using the response template.

Matching Rules

• Keyword-Based: Matching is based on predefined keywords in each strategy
• Case-Insensitive: Keyword matching is case-insensitive
• Priority: Strategies are evaluated in order; first match wins
• Default Strategy: If no strategy matches, the "default" strategy is used

Available Strategies

1. Shopping Search

• Intent: Search for products
• Keywords: search, find, look for, show
• Tool: shopping.search_products
• Parameter: Extracts keywords from user input using pattern search\s+(.+)
• Example: "search smartphone" → searches for products matching "smartphone"

2. Shopping Purchase

• Intent: Purchase products
• Keywords: purchase, buy, order
• Tool: shopping.purchase_product
• Parameters:
  • productId: Extracted using pattern purchase\s+(\w+)
  • quantity: Defaults to "1"
• Example: "purchase PROD-005" → purchases product with ID PROD-005

3. Greeting

• Intent: Greet the user
• Keywords: hello, hi, hey
• Tool: None (no-tool: true)
• Response: Fixed greeting message

4. Default

• Intent: Handle unrecognized requests
• Keywords: None (catch-all)
• Tool: None (no-tool: true)
• Response: Error message asking user to try again

Matching Algorithm

The Mock LLM uses a word-based matching algorithm:

1. Keyword Extraction: Extracts keywords from the user's natural language request
2. Product Search: Searches for products where ANY keyword matches the product name or description
3. Category Filter: Optionally filters by product category

Matching Rules

• Case-Insensitive: Matching is case-insensitive (e.g., "phone" matches "Smartphone")
• Partial Match: Keywords can match partial words (e.g., "phone" matches "Smartphone")
• Any Keyword Match: If multiple keywords are provided, a product matches if ANY keyword matches
• Name & Description: Matching is performed against both product name and description

Example

User Request: "I want to buy a smartphone"

Keywords Extracted: ["want", "buy", "smartphone"]

Matching Products:

• ✅ "Smartphone X" (matches "smartphone")
• ✅ "Laptop Pro" (no match)
• ✅ "Wireless Headphones" (no match)

Available Products

The sample project includes 12 default products across 3 categories:

Clothing (4 products)

ID	Name	Description	Price
PROD-001	Down Jacket A	Excellent warmth, suitable for daily commute	¥599.00
PROD-002	Wool Coat B	Fashionable and versatile, suitable for business	¥899.00
PROD-003	Cotton Jacket C	Lightweight and comfortable, suitable for casual wear	¥399.00
PROD-004	Wool Sweater D	Soft and comfortable, warm and breathable	¥299.00

Electronics (4 products)

ID	Name	Description	Price
PROD-005	Smartphone X	Latest model with advanced features	¥4,999.00
PROD-006	Laptop Pro	High performance for professionals	¥8,999.00
PROD-007	Wireless Headphones	Noise cancelling, premium sound	¥1,299.00
PROD-008	Smart Watch	Fitness tracking and notifications	¥1,999.00

Books (4 products)

ID	Name	Description	Price
PROD-009	Java Programming Guide	Comprehensive Java tutorial	¥89.00
PROD-010	Design Patterns	Classic software design patterns	¥79.00
PROD-011	Clean Code	Writing maintainable code	¥69.00
PROD-012	System Design Interview	System design preparation	¥99.00

Sample Queries

Here are some example queries and their expected results based on the configured strategies:

Query	Matched Strategy	Keywords Matched	Expected Result
"search smartphone"	shopping-search	search	Searches for products matching "smartphone"
"find laptop"	shopping-search	find	Searches for products matching "laptop"
"show me books"	shopping-search	show	Searches for products matching "books"
"purchase PROD-005"	shopping-purchase	purchase	Purchases product with ID PROD-005
"buy PROD-001"	shopping-purchase	buy	Purchases product with ID PROD-001
"hello"	greeting	hello	Returns greeting message
"hi there"	greeting	hi	Returns greeting message
"random text"	default	none	Returns error message

Tips for Testing

1. Use Strategy Keywords: Start your query with one of the strategy keywords (search, find, show, purchase, buy, order)
2. For Search: Use "search [keyword]" or "find [keyword]" to search for products
3. For Purchase: Use "purchase [productId]" or "buy [productId]" to purchase a specific product
4. Product IDs: Use the product IDs from the table above (PROD-001 to PROD-012) for purchases
5. Try Greetings: Use "hello", "hi", or "hey" to test the greeting strategy

Limitations

• No Real AI: Mock LLM does not use actual AI models
• Simple Matching: Uses basic keyword matching, not semantic understanding
• Fixed Dataset: Only searches the predefined 12 products
• No Context: Does not maintain conversation context

For Real AI Experience

To use a real AI model for more natural interactions, see Using QwenCode in the Quick Start Guide.

---

Document Version: 1.0.0
Last Updated: 2026-02-13
Maintainer: Open Agent Auth Team