Overview

This project is a context-aware chatbot built using LangChain JS, a framework that helps developers create reasoning applications by abstracting common functionalities. The chatbot leverages embeddings, vector stores, and chains to process and respond to user queries effectively.

Features

1. Embeddings

• Embeddings are mathematical representations of text that AI understands. They are used to plot relationships between words in vector space.
• This project uses embeddings to encode text and store them in a vector database (Supabase).

2. Vector Store Integration
   • Supabase is used as the vector store to store and retrieve embeddings efficiently.
   • The project includes functionality to store documents and perform similarity searches using embeddings.

3. Templates and Prompts
   • Templates are used to create dynamic prompts for the chatbot.
   • Prompts are designed to guide the chatbot in generating context-aware responses.

4. Chains
   • Chains are used to combine multiple steps in the reasoning process.
   • The project uses RunnableSequence to create a pipeline of operations, such as generating standalone questions, retrieving relevant documents, and generating answers.

5. .pipe() Method
   • The .pipe() method is used to connect different components, such as prompts, LLMs, and output parsers, into a seamless workflow.

6. Retrieval from Vector Store
   • The chatbot retrieves relevant documents from the vector store based on user queries.
   • The retrieved documents are used as context for generating accurate responses.

7. LLM Integration
   • The project integrates a language model (LLM) to process prompts and generate responses.
   • The LLM is configured with a specific model and API key.

Project Workflow

1. Text Preprocessing
   • Input text is split into smaller chunks using a text splitter.
   • Each chunk is embedded and stored in the vector store.

2. Document Storage
   • Documents are stored in the Supabase vector store with metadata for efficient retrieval.

3. Query Processing
   • User queries are converted into standalone questions using a prompt template.
   • Relevant documents are retrieved from the vector store based on the query.

4. Response Generation
   • The chatbot generates responses using the retrieved context and a predefined answer template.
   • The response is processed and returned to the user.

How to Run the Project

1. Clone the repository:

   git clone <repository-url>
   cd Context_Aware_Chatbot_With_LangChain_JS
   

2. Install Dependencies
   Install all the required dependencies for the project using the following command:

   npm install

3. Set Up Environment Variables
   Create a .env file in the root directory of the project and add the following environment variables:

   SB_API_KEY=<your-supabase-api-key>
   SB_API_URL=<your-supabase-url>
   OPENAI_API_KEY=<your-openai-api-key>
   
   ** Replace <your-supabase-api-key>, <your-supabase-url>, and <your-openai-api-key> with your actual API keys and URLs.
   

4. Ensure Supabase Database Setup

• Verify that the documents table exists in your Supabase database with the following schema:

create table if not exists documents (
    id uuid primary key default gen_random_uuid(),
    text text,
    embedding vector(1536) -- Adjust the dimension to match your embeddings
);

• Create the match_documents function in your Supabase database to enable similarity searches:

create or replace function public.match_documents(
    query_embedding vector(768), -- Adjust the dimension to match your embeddings
    similarity_threshold float, 
    match_count int
) 
returns table(
    id uuid,
    text text,
    similarity float
)
language plpgsql
as $$
begin
    return query
    select
        id,
        text,
        1 - (documents.embedding <=> query_embedding) as similarity
    from
        documents
    where
        1 - (documents.embedding <=> query_embedding) > similarity_threshold
    order by
        similarity desc
    limit match_count;
end;
$$;

• Ensure that the documents table and match_documents function are correctly set up in your Supabase database to enable the chatbot to perform similarity searches.