Enterprise Search

Enterprise Search is a Retrieval-Augmented Generation (RAG) system designed for efficient local information retrieval on document collections. It uses vector search techniques with language models to provide context-aware answers to your queries.

Enterprise Search is designed to be dynamic and adaptable to fit in both development and production workflows. It offers a RestFul API for indexing and querying over document collections, making it ideal for businesses and developers seeking to deploy the question-answering solutions in their infrastructure.

High level overview of user interaction with the Enterprise Search system

📚 Table of Contents

• Key Features
• Prerequisites
• Quick Start
• Configuration
• API
• Testing
• Evaluation
• Deployment
• UI
• License
• Acknowledgements

🚀 Key Features

Enterprise Search - Key Features

🛠️ Prerequisites

Before setting up Enterprise Search, ensure you have:

• Python 3.9 or higher
• Docker and Docker Compose
• CUDA 11 or higher (for GPU acceleration)

Note: Depending on your Docker Compose version, you may need to use docker-compose (with a hyphen) instead of docker compose.

⚙️ Configuration

1. Rename .env.example to .env and update the values to match your setup.

2. Update the configuration in config/config.dev.yaml. Default settings are defined in llamasearch/settings.py:

• application: Application settings
• vector_store_config: Qdrant settings for vector storage
• qdrant_client_config: Qdrant client connection settings
• redis_config: Redis settings for document store and cache
• embedding: Embedding model configuration (uses model from HuggingFace)
• llm: Language model configuration (uses model from ollama/openai)
• reranker: Reranker model settings (uses model from HuggingFace)

3. Setup LLM of your choice.

Open-Source Option: Ollama

1. Run Ollama docker:: Use the docker/docker-compose-ollama.yml file to run Ollama:

docker-compose -f docker/docker-compose-ollama.yml up -d

The docker container will pull models mentioned in the config/config.dev.yaml on startup. It may take few minutes to download the models. Check ollama docker logs for progress.

External Providers Option: OpenAI

To use OpenAI's proprietary models, set OPENAI_API_KEY in .env file.

1. Set up OpenAI API key: Export your OpenAI API key:

export OPENAI_API_KEY=your_api_key_here

2. Configure for OpenAI:: Update the config/config.dev.yaml file to use an OpenAI model, set use_openai flag to True. Check the openai models list.

🚀 Quick Start

Option 1: Jupyter Notebook

1. Clone the repository and set up the environment:

conda create --name es_env python=3.9
conda activate es_env
pip install -r requirements.txt

2. Start Jupyter Notebook:

pip install jupyter
jupyter notebook

3. Open the quick_start.ipynb file in your browser and follow the step-by-step instructions to set up and test the pipeline.

Option 2: Command Line Interface

1. Set up the environment (same as above)

2. Configure the application:

• Modify config/config.dev.yaml to match your setup.

4. Start the redis and qdrant services:

docker-compose -f docker/docker-compose.yml up -d redis qdrant

5. Run the pipeline::

python -m llamasearch.pipeline

The pipeline loads documents from application->data_path defined in config file, processes and indexes them on startup. Enter your query when prompted. Results will be displayed in the terminal.

🌐 API

We provide a RESTful API for document indexing, querying, and management. Follow steps to test the pipeline and backend server (API) using curl locally.

Default API settings are defined in llamasearch/api/core/config.py. These settings can be customized using environment variables defined in .env file in the project root.

Important: When deploying to production, ensure you set appropriate values related to server, authentication.

1. Build the Docker Image: Run the following command to build the Docker image.

docker build -t es:latest -f docker/Dockerfile .

2. Authentication: Update FIREBASE_CREDENTIALS_PATH to point to your firebase credentials file in .env file for user authentication. Refer to Firebase README for instructions.

Note: Currently, we only support testing API endpoints with authentication enabled. A firebase account is required to test the API endpoints.

3. Setup LLM: Setup the LLM of your choice (if you haven't already) as mentioned in the configuration section. Ensure the LLM service is running.

4. Run the docker image: Adjust docker mount points in the docker/docker-compose.yml file to point to match your local setup. It will run the API server on port 8010 by default.

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

5. Test API: For API usage examples, including request and response formats, curl request examples and more, please refer to our API Documentation.

🧪 Testing

We use pytest for testing. To run the test suite:

1. Ensure you're in the project root directory.

2. Start the API server as stated in the API section.

3. In another terminal, set up the Python path:

export PYTHONPATH=$PYTHONPATH:$(pwd)

4. Run the tests:

pytest

For more detailed testing instructions, including how to run specific tests, please refer to our Testing Guide.

📊 Evaluation

Enterprise Search includes a robust evaluation module based on DeepEval to assess the RAG pipeline's performance:

• 🧪 Synthetic dataset generation for simulating your documents
• 📈 Industry standard metrics including faithfulness, relevancy, and coherence

For step-by-step instructions on running evaluations, see our Evaluation Guide.

🚀 Deployment

Enterprise Search can be deployed using Kubernetes and Helm. Here's a high-level overview of the deployment process:

1. Build and push the Docker image to your docker registry:

docker build -t es:latest .
docker push es:latest

2. Configure your Kubernetes cluster and ensure kubectl is set up correctly.

3. Update the k8s/values.yaml file with your configuration settings (namespace, service names, etc.).

4. Deploy using Helm:

cd k8s/
helm install enterprise-search . --values values.yaml

5. Monitor the deployment:

kubectl get pods,svc -n {{YOUR_NAMESPACE}}

For detailed deployment instructions, please refer to our Deployment Guide.

🖥️ UI

We provide an experimental frontend application built with Next.js 14 (app router) to interact with the Enterprise Search API.

For detailed setup instructions and usage guidelines, refer to the README.

📄 License

This project is licensed under the SOFTWARE LICENCE AGREEMENT - see the LICENSE file for details.

🙏 Acknowledgements

Enterprise Search project is built on top of valuable open source projects. We'd like to acknowledge the following projects and their contributors:

• LlamaIndex for a stable foundation for RAG capabilities with wide array of integrations
• Deepeval for the RAG evaluation framework
• Qdrant for the vector database functionality
• FastAPI for the high-performance web framework
• Ollama for local LLM inference
• Redis for caching and document storage
• Docker for containerization
• Kubernetes for orchestration