Refactor the Infoslicer Activity to generate plain-language summaries

[walterbender]

Ticket Contents

Description

The Infoslicer Activity is designed to help teachers extract content from the Wikipedia in order to create lesson plans. This is currently a manual, extractive process. It is well suited to generative AI. The goal would be to have a teacher type in a theme for a lesson and have the AI create a simple lesson plan, which the teacher can then edit.

The biggest challenge to summarization using generative AI is hallucinations. A work-around for this is to include a validation step that surfaces evidence (or lack of evidence) for each assertion in the lesson plan. This will introduce some workflow and UX challenges.

Goals & Mid-Point Milestone

Goals

• [Research suitable LLM]
• [Backend development]
• [Refactor Infoslicer]
• [Goals Achieved By Mid-point Milestone]
• [Research suitable LLM]
• [Backend development]

Setup/Installation

No response

Expected Outcome

A refactoring of Infoslicer that can produce summaries of Wikipedia articles at a specified grade level.

Acceptance Criteria

• backend server
• integration with Infoslicer
• grade-level summaries

Implementation Details

• docker image for LLM
• potential rewrite/simplification of Infoslicer activity

Mockups/Wireframes

No response

Product Name

Infoslicer Activity

Organisation Name

Sugar Labs

Domain

⁠Education

Tech Skills Needed

Artificial Intelligence, Docker, Python

Mentor(s)

@chimosky @Wa

Category

Backend, Frontend, AI