Search-AI-Demo

Overview

This project is a Spring Boot application that demonstrates the use of AI models for semantic search and chat functionalities. The project is divided into two main parts:

1. SemanticSearchApplication: The main application that sets up the Spring Boot environment and mocking the necessary services for semantic search.
2. SpringAiTests: A set of tests to demo some capabilities of the AI models and services integrated into the application.

SemanticSearchApplication

A Spring Boot application that implements AI-powered semantic search for users within an organization.

Key Components

1. SearchController: Handles API requests for user searches
2. SearchService: Core logic for processing natural language queries
3. OllamaConfig: Sets up AI model and function callbacks
4. Repositories: Simulate databases for users, projects, teams, and mentions
5. Converters: Transform AI responses into structured data

Key Features

• Natural language query processing
• Context-aware search (org, team, user)
• Integration with Ollama for AI functionality
• Flexible data retrieval via function callbacks

SpringAiTests

The SpringAiTests class contains various methods to demo the functionality of the AI models and VectorStore with help of PGVector and Llama3.1.

• tellMeJoke: Sends a simple user prompt asking for telling a joke.

• tellMeJokeSystemPrompt: Asks for a joke in different "voices" (e.g., Picasso, Musk, Shakespeare) using a system prompt.

• structuredOutput: Retrieves a structured output for a given prompt.

• structuredOutputList: Retrieves a structured output for a given prompt, maps it to a list.

• noContext: Sends a prompt without additional context.

• stuffPrompt: Sends a prompt with extra contextual information.

• preLoadData: Preloads data and write into PGVector, which preparing system for subsequent question-answering method.

• purposeQuestionWIthPreLoadedData: Asks a question that needs preloaded data and evaluates the relevancy of the response. (RAG)

• conversationMemory: Tests chat memory by asking the user’s name and recalling it in subsequent interactions.

• multiModality: Sends a prompt involving an image and asks for an explanation of its contents.

Running the Application

Prerequisites

• Java 11 or higher
• Maven
• Docker

Running the Application Locally

1. Clone the repository:

   git clone https://github.com/your-repo/Search-AI-Demo.git
   cd Search-AI-Demo

2. Running Required LLMs and Containers:

   • Navigate to the Docker-Script directory:

   cd Docker-Script

   • Run the Docker Compose file:

   docker-compose -f docker-compose-pgvector.yml up -d

3. Running Required LLMs:

   • Follow the instructions on the Llama3.1 GitHub page to download and set up the models locally and after that:

     ollama pull llama3.1
     ollama pull llava