Sensei Search

Sensei Search is an AI-powered answer engine.

📸 Screenshots

Light Mode

Dark Mode

💡 Insights from Utilizing Open Source LLMs

The key takeaways and experiences of working with open source Large Language Models are summarized in a detailed discussion. For more information, you can read the full discussion on Reddit:

• Building an Open Source Perplexity AI with Open Source LLMs - Reddit Post

🛠️ Tech Stack

Sensei Search is built using the following technologies:

• Frontend: Next.js, Tailwind CSS
• Backend: FastAPI, OpenAI client
• LLMs: Command-R, Qwen-2-72b-instruct, WizardLM-2 8x22B, Claude Haiku, GPT-3.5-turbo
• Search: SearxNG, Bing
• Memory: Redis
• Deployment: AWS, Paka

🏃‍♂️ How to Run Sensei Search

You can run Sensei Search either locally on your machine or in the cloud.

Running Locally

Follow these steps to run Sensei Search locally:

1. Prepare the backend environment:

   cd sensei_root_folder/backend/
   mv .env.development.example .env.development

   Edit .env.development as needed. The example environment assumes you run models through Ollama. Make sure you have reasonably good GPUs to run the command-r/Qwen-2-72b-instruct/WizardLM-2 8x22B model.

2. No need to do anything for the frontend.

3. Run the app with the following command:

   cd sensei_root_folder/
   docker compose up

4. Open your browser and go to http://localhost:3000

Running in the Cloud

We deploy the app to AWS using paka. Please note that the models require GPU instances to run.

Before you start, make sure you have:

• An AWS account
• Requested GPU quota in your AWS account

The configuration for the cluster is located in the cluster.yaml file. You'll need to replace the HF_TOKEN value in cluster.yaml with your own Hugging Face token. This is necessary because the mistral-7b and command-r models require your account to have accepted their terms and conditions.

Follow these steps to run Sensei Search in the cloud:

1. Install paka:

   pip install paka

2. Provision the cluster in AWS:

   make provision-prod

3. Deploy the backend:

   make deploy-backend

4. Deploy the frontend:

   make deploy-frontend

5. Get the URL of the frontend:

   paka function list

6. Open the URL in your browser.