Question-Answer Bot with FastAPI, Elasticsearch, and OpenAI Embeddings

Overview

This project implements a Question-Answer (QA) bot using FastAPI and Elasticsearch. It is designed to receive questions and provide relevant answers by leveraging the power of Elasticsearch for efficient data retrieval and OpenAI's embedding model for semantic understanding.

Prerequisites

• Python 3.x
• Docker
• Elasticsearch

Setup Instructions

Setting up the Virtual Environment

1. Ensure Python 3.x is installed on your system.
2. Follow the guide here to set up a Python virtual environment.
3. Activate the virtual environment.
4. Install the required dependencies by running pip install -r requirements.txt (ensure you have a requirements.txt file with all necessary packages).

Running Elasticsearch with Docker

1. Install Docker on your machine if you haven't already.
2. Start Elasticsearch using Docker by executing the following command in the terminal:

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

This will set up Elasticsearch to run on localhost:9200.

Running the Application

1. Start the FastAPI server by running all the cells in juypter notebook.
2. The API will be available at http://localhost:5001.
3. Elasticsearch can be accessed at http://localhost:9200.

API Endpoints

• POST /train/: Add a question-answer pair to the system.
• POST /ask/: Retrieve answers for a given query.
• GET /health/: Check the health status of the API.

High-Level Architecture

The application uses FastAPI for handling HTTP requests. Questions and answers are indexed in Elasticsearch with their respective vector representations. OpenAI's embedding model is used to convert textual data into vectors for semantic search. When a query is made, the system finds the most relevant answers using cosine similarity in Elasticsearch.