Custom data chatbot

Building a custom Q&A chatbot with supabase and openAI model

Supabase - an open-source database infrastructure built on PostgreSQL.

OpenAI Models - models used here are gpt-4 and text-embedding-ada-002

A Supabase and OpenAI system is built, that allows to call an API endpoint with a question, and then generate an answer based on the custom data provided. This is done by using PostgreSQL vector storage, Supabase Edge functions, and OpenAI Embeddings. A supabase and OpenAI account is required.

Create a .txt file with custom data

Data about a fake person named Anitha is stored in .txt file, it's then turned into an embedding using the model text-embedding-ada-002.Embeddings are stored in supabase vector.

id	content	embedding
1	Anitha grew up in a small town in Kerala.	[0.019696854, 0.00064515544, 2.64515544]
2	After excelling in math and science classes in high school.	[0.0019600854, 0.01064515544, 0.64515544]

Define an API end point

A post endpoint is defined to receive questions in the request body. The questions are then converted into embeddings.

Supabase Edge function gets called with question and use match_documents function

match_documents returns an array of matching documents to the question.Then loop through the array of documents, format them for chatgpt promt.gpt-4 model is used.

  const completionResponse = await openai.chat.completions.create({
    messages:[{
        role: "system",
        content: `You are a happy and helpful assistant named Poppy.Who answers qustions based on the context sections
        ${matching_documents_array}`,
      },
    { role: "user", content: "<question from user>? },],
     model: "gpt-4",
    })

Finally return the response from the model to user through a response

Source code

https://github.com/AngelMathew/chatbotApi.git