ramamimu/chatbot-api

As final project for my Bachelor, i create RAG (Retrieval Augmented Generation) as back-end service which combine Retrieval and Generation to get information from bunch of documents. The Embedding use model from Huggingface and Text Generation from OpenAI

Chatbot API

how to run manually

if run manually, only supported on unix

• create virtual environment

  python3 -m venv venv
  
  source venv/bin/activate
  

• install packages

  Highly recommend to install package manually as put in installer.txt

  or

  pip3 install -r requirements.txt
  

• create .env such example.env file

• setup database

  setup the database as put in .env file. then create database name chatbot

  CREATE DATABASE chatbot
  

  or you can simply use docker compose of postgres in compose/postgres.yaml

• Apply the Alembic migrations

  make migrate-checkout r=head
  

• download embeeding model

  python3 model.download.py
  

• embed basic knowledge for vectorstore db

  python3 embed.init.py
  

• Configure preprocessing file as mention at preprocessing section (optional)

• run the app

  for development

  python3 app.py
  

  for deployment testing

  fastapi run app.py
  

how to run using docker

1. create .env such example.env file

2. Configure preprocessing file as mention at preprocessing section (optional)

3. run docker compose command

docker compose build --no-cache

docker compose up -d

-d means running as daemon

exposing port 5001 as default

sometimes, the code was error. Keep build the image untill get succeed then compose up

how to add preprocessing file

you need to configure of three things:

1. create folder name by following rule documents/preprocessing-<your custom name>

2. provide .pdf file in your directory as downloadable file later on

3. provide .txt file in your directory which contains .pdf file content as chatbot knowledge due to not all .pdf file is readable

the program will automatically recognize as preprocessing stuff and will be loaded when it get starts.

documentation

Documentation able to see on url/docs. It is generated automatically by fastapi. It also provides API playground.