Skip to main content Skip to navigation

Kwame for Science: An AI Teaching Assistant Based on Sentence-BERT for Science Education in West Africa

Project Overview

The document discusses the significant role of generative AI in education, highlighting the implementation of Kwame for Science, an AI teaching assistant aimed at enhancing science education in West Africa. By leveraging a Sentence-BERT model, Kwame provides instant responses to student inquiries aligned with the Integrated Science curriculum for the West African Senior Secondary Certificate Examination (WASSCE). The initial deployment of this AI tool demonstrated a high accuracy rate, which underscores its effectiveness in addressing the challenges posed by a high student-to-teacher ratio and the need for improved access to quality education. The initiative exemplifies how generative AI can support personalized learning experiences and facilitate educational equity, particularly in regions facing resource constraints. Overall, the findings suggest that integrating AI technologies like Kwame can significantly enhance student engagement and learning outcomes in science education, paving the way for further innovations in the educational landscape.

Key Applications

Kwame for Science

Context: Science education for senior high school students in West Africa

Implementation: Deployed as a web app using Sentence-BERT for question answering, displaying answers with confidence scores and related past exam questions.

Outcomes: Top 3 accuracy of 87.5% in providing useful answers; scalable, cost-effective education delivery.

Challenges: Issues with typos in scientific terms, out-of-scope questions, and incomplete answers due to dataset limitations.

Implementation Barriers

Technical barrier

Challenges related to the accuracy of the AI in understanding and answering questions due to spelling errors, dataset limitations, and the need for fine-tuning.

Proposed Solutions: Future plans to fine-tune the SBERT model with real-world data to improve accuracy.

Resource barrier

Difficulty in obtaining permission to use proprietary educational materials for training the AI.

Proposed Solutions: Utilized open-source textbooks and datasets to curate training data.

Project Team

George Boateng

Researcher

Samuel John

Researcher

Andrew Glago

Researcher

Samuel Boateng

Researcher

Victor Kumbol

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: George Boateng, Samuel John, Andrew Glago, Samuel Boateng, Victor Kumbol

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies