SQLite Retrieval Augmented Generation (RAG) and Vector Search With C# and Ollama

This project demonstrates a simple movie recommendation system using the Semantic Kernel and the SQLite Vector Store connector. It leverages vector embeddings to store and query movie data based on textual descriptions and performs similarity searches to provide relevant recommendations.

Features

• SQLite Vector Store: Uses SQLite with the vector extension (vec0) to store and query movie embeddings.
• Embedding Generation: Generates vector embeddings for movie descriptions using the OllamaEmbeddingGenerator.
• Vectorized Search: Supports similarity-based search to recommend movies based on a user-provided query.
• C# Implementation: Fully implemented in C# using the Microsoft Semantic Kernel library.

Prerequisites

1. .NET 9.0 or later.

2. A SQLite build with the vec0 module (vector extension) enabled.

3. NuGet Packages:

   • Microsoft.Data.Sqlite
   • Microsoft.Extensions.AI.Ollama
   • Microsoft.Extensions.AI
   • Microsoft.Extensions.VectorData
   • Microsoft.SemanticKernel.Connectors.Sqlite

4. Ollama Embedding Generator running locally (e.g., at http://localhost:11434/).

Installing and Using MiniLM-L6-v2 in Ollama

To install and use the MiniLM-L6-v2 (or all-MiniLM-L6-v2) model in Ollama, follow these steps:

1. Ensure Ollama is Installed

   • If Ollama is not installed, download and install it from the Ollama website.
   • Ensure you have the latest version (v0.1.26 or newer) to support MiniLM-L6-v2.

2. Verify Model Availability

   • Ollama provides built-in support for embedding models, including MiniLM-L6-v2. The model might be referred to as all-minilm:l6-v2 in Ollama.
   • Visit Ollama's model library or search using their CLI/API to confirm:

     ollama list

3. Install the Model

   • Use Ollama's CLI command to download and install the model:

     ollama pull all-minilm:l6-v2

     This command will fetch the all-MiniLM-L6-v2 model into your Ollama environment.

Setup Instructions

1. Clone the Repository

git clone <repository-url>
cd <repository-folder>

2. Install Dependencies

Restore NuGet packages:

dotnet restore

3. Ensure SQLite Vector Support

• Download and install a SQLite version that includes the vec0 module.SQLite doesn't support vector search out-of-the-box. The SQLite extension should be loaded first to enable vector search capability. The current implementation of the SQLite connector is compatible with the sqlite-vec vector search extension.

4. Configure the Project

• SQLite doesn't support vector search out-of-the-box. The SQLite extension should be loaded first to enable vector search capability. The current implementation of the SQLite connector is compatible with the sqlite-vec vector search extension.
• Ensure the SQLite database file (movies.db) exists, or it will be automatically created on the first run. Update the database connection string in the code if necessary.

5. Run the Application

Build and run the application:

dotnet run

How It Works

1. Movie Data Initialization:

   • A list of movies with titles and descriptions is initialized.
   • Vector embeddings are generated for each movie description.

2. Data Storage:

   • Embeddings and movie metadata are stored in the SQLite vector store.

3. Query and Recommendations:

   • The user provides a query (e.g., "A family-friendly movie with ogres and dragons").
   • The query is converted to an embedding and searched against the stored movie vectors.
   • The top matching movies are retrieved and displayed.

Example Output

Example query: "A family-friendly movie with ogres and dragons."

Sample result:

Title: Shrek
Description: Shrek is an animated film that tells the story of an ogre named Shrek who embarks on a quest to rescue Princess Fiona from a dragon and bring her back to the kingdom of Duloc.
Score: 0.95

Troubleshooting

1. SQLite Error: 'no such module: vec0':

   • Ensure you are using a SQLite version with vector extension support.
   • Specify the full path to the vec0 module using connection.LoadExtension("path/to/vec0");.

2. Embedding Generator Issues:

   • Verify the OllamaEmbeddingGenerator is running and accessible at the specified URL.

Contributing

Feel free to contribute to this project by submitting issues or pull requests. Suggestions and improvements are always welcome!

License

This project is licensed under the MIT License.