https://github.com/mahmoudajawad/conversational_kb

A PoC to created console-based chat interface to knowledge based using OpenAI gpt-3.5-turbe model
https://github.com/mahmoudajawad/conversational_kb

ai chatbot chatgpt conversational conversational-ai gpt-3

Last synced: 11 months ago
JSON representation

A PoC to created console-based chat interface to knowledge based using OpenAI gpt-3.5-turbe model

• Host: GitHub
• URL: https://github.com/mahmoudajawad/conversational_kb
• Owner: mahmoudajawad
• License: agpl-3.0
• Created: 2023-03-22T09:34:46.000Z (over 3 years ago)
• Default Branch: master
• Last Pushed: 2023-03-25T05:01:09.000Z (over 3 years ago)
• Last Synced: 2025-07-06T12:08:39.363Z (about 1 year ago)
• Topics: ai, chatbot, chatgpt, conversational, conversational-ai, gpt-3
• Language: Python
• Homepage:
• Size: 427 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Conversation KB

A PoC to created console-based chat interface to knowledge based using OpenAI `gpt-3.5-turbe` model.

Knowledge is provided as titled paragraphs in plain text format in `articles` directory, that is then vectorised into embeddings and saved to `embeddings` directory. With every question asked, the question is also vectorised and its embeddings are matched against the paragraphs embeddings to find three top most matches that are then appended to a `system` prompt for `gpt-3.5-turbo` model. If the question isn't the first, last top match is also appended to `system` prompt to enable follow-up questions.

## How to Run

- Add your own articles paragraphs to `articles` directory, or use provided samples from `samples` directory:

  - Title every paragraph.

  - Use mid-size paragraphs only.

  - embeddings are calculated when no embeddings matching file exist, which means if you modify a file in `articles` directory that its embeddings are already calculated, you will have to manually remove the matching `embeddings` directory file to force re-calculation.

- Create a `venv`: `python -m venv venv`.

- Install dependencie: `python -m pip install -r requirements.txt`.

- Run: `python .`

The project is set to emit all kind of logs, as this is a PoC that is meant to understand how such a simple approach to conversation knowledge-base would play, however, you can manually set logging level in first line of `main` function.