Teaching AI Through Narrative Data: A Practical Framework for Data Science and Retrieval-Augmented Generation

Read full text Summarise with VIRAT
Authors: Charlie Dey, Susan Lindsey
Cited by: 0
References: 5

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Artificial intelligence (AI) and machine learning (ML) education has traditionally been split between technical model-building and data literacy.While these skills are often taught separately, the emergence of large language models (LLMs) offers an opportunity to unify them through narrative-driven, human-readable data transformation.This approach enables learners to query structured data using natural language while still engaging deeply with the underlying analytical processes.We present a hands-on educational framework-debuting at the 2025 Big Data School in Costa Rica-that grounds AI learning in real-world data by transforming a single, richly structured dataset into narrative text that LLMs can ingest and reason over.Using the Austin Real-Time Traffic Incident Reports, participants apply core data science techniques-classification, clustering, regression, and forecasting-before extending their work into Retrieval-Augmented Generation (RAG) pipelines.A key design element is the careful selection of a dataset that supports multiple analytical tasks, contains time-series and categorical diversity, and can be effectively compressed into natural-language summaries without losing critical meaning.This continuity of dataset enhances comprehension, data intuition, and knowledge transfer without the cognitive effort of switching contexts.By converting tabular data into concise, human-readable narratives, learners bridge traditional analytics and AI-enhanced insight generation.Outcomes include skill-building in Python-based ML pipelines, embedding and vector retrieval, and critical reflection on model interpretability, hallucination risk, and accessibility for non-technical users.This narrative-driven, single-dataset strategy-combined with explicit dataset selection principles-supports scalable adoption for high-performance computing and AI pedagogy in both academic and professional environments.

Fields of Study & Subjects

Statistics Education and Methodologies Teaching and Learning Programming Online Learning and Analytics

Publication details

Journal / source
The Journal of Computational Science Education
Volume
17
Issue
1
Pages
50–56
Publication date
1 March 2026
Language
EN