https://github.com/svuit/ai-assistant

https://github.com/svuit/ai-assistant

Last synced: 11 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/svuit/ai-assistant
• Owner: SVUIT
• Created: 2024-07-06T02:29:59.000Z (7 months ago)
• Default Branch: main
• Last Pushed: 2024-10-22T17:21:28.000Z (3 months ago)
• Last Synced: 2024-11-11T07:46:27.496Z (2 months ago)
• Language: HTML
• Size: 601 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# AI assistant for SVUIT-MMTT

## Front end

### 1. Initialize Widget

- Event: User visits website.

- Action:

    + Widget appears in a fixed position in the lower right corner.

    + Display a chat icon and the text Ask AI.

### 2. User Initiates Conversation

- Event: User clicks on widget.

- Action:

    + displays the chat frame and notes about the chatbot's effects.

    + Content outside the chat frame is blurred.

      

### 3. User Enters Message

- Event: User enters text into the chat box.

- Action:

  + User's message is sent (Displayed on the right side of the chat frame).

  + Bot processes messages and responds appropriately (Displayed on the left side of the chat frame).

### 4. Reply Bot

- Event: The bot has finished processing the user's message.

- Action:

  + Bot sends reply messages, which can include:

    + Text: Direct answers to user questions.

    + Markdown: Display information as Markdown file.

    + Button:

    	+ Copy: let users copy the answer from the bot.

    	+ Refresh: so that the user wants the bot to answer the previously asked question again.

    

### 5. Other Interactions

- Users review chat history: Allows users to review previous conversations by scrolling up.

### 6. End the Conversation

- Event: User closes the tab.

- Action:

  + Widget shrinks back to its original state and deletes all previous chat content.

## Back end

- Idea: Building a chatbot using the Gemini base model and the LangChain LLM framework.  

- Language: python

-  Training data: content from courses of University of Information Technology, Viet Nam National University , Ho Chi Minh City. 

- Steps:

	+  Using DirectoryLoader of langchain_community to load data 

		+ LangChain's DirectoryLoader provides functionality for reading various file types from disk into LangChain Document objects. In this project, the data includes three types: pdf, docx, md.

	+ Split documents into chunks using RecursiveCharacterTextSplitter.

	+ Using model embedding-001 to embed the contents of each document  and insert these embeddings into the Chroma vector store.

	+ Creating a vector retriever from a vectorstore.

	+ Create a retriever from vector retriever and chat history.

	+ Build the prompt for the question and answer.

	+ Create a chain using the retriever, prompt, and the gemini-1.5-flash model.

	+ Build an API with Flask to generate answers.

## Infra

- Encapsolution: The frontend and backend code is encapsoluted to run on Google Cloud enviroment using a Dockerfile

- Container management: Containers are spawned using Docker Compose. When a user access to web, a container that holds an AI service is created for them. This helps ensure per-user containerization    

- Socket Management: Sockets are used to maintain connections between the user and the server proxy. Each user has a unique socket id to differentiate them. Additionally, sockets enable faster communication from user to container that holding the AI service compared to http    

- Proxy server: The proxy server is responsible for creating and closing sockets, as well as managing the creation and deletion of service containers. It also ensures per-user containerization.

- Add gemini api key to .env file

## Run the project

### Run frontend code 

1. Install Docker 

2. Navigate to the folder containing the Dockerfile.

3. Build the Docker image: "sudo docker build -t [ your-docker-image-name ] ."

4. Run the Docker container: "sudo docker run -p [ host-port ]:4000 docker-image-name

### Run backend code 

1. Install Docker

2. Navigate to the folder containing "compose.yml"

3. Run the Docker Compose command: "sudo docker compose up --build"