مهندس الكلمة - The Word Engineer

ALLaM Challenge - SDAIA

August - November 2024

This is our submission for ALLaM Challenge 2024, hosted by the Saudi Data & AI Authority (SDAIA).

Presentation Slides

Authors

• Abdullah Alwakeel
• Abdulrahman Alshahrani

Project Description

Our project aims to improve the quality of AI Arabic poetry generation by using chain-of-thought and trial-and-error techniques.

Chain of Thought

With chain-of-thought, the model is able to set a plan for its poem before generating it, and also iteratively critic its own generation. This technique reduces the amount of linguistic errors and increases the response's relevancy to the user prompt.

Trial and Error

Powered with poetry analysis tools, the model is also able to backtrack on incorrect generations and retry to ensure the generated poems are correct linguistically and follow Arabic poetry rules.

Usage

- Backend (Python Flask)

Install dependencies

pip install -r requirements.txt

Running the backend API server

python website.py

Alternatively, you can run a CLI (Command-Line Interface) version

python main.py

- Frontend (Vite + React + TS)

Install dependencies

cd allam-website && npm i

Running the frontend

npm run build