Slides + next module
Reopen the full deck for interactive elements, then move to the data lakehouse module to continue building your operational AI foundation.
Workshop recap companion
The intersection between AI and data science
Artificial intelligence and data science are transforming research analytics. This session unpacks how artificial intelligence and data science intersect to support actionable applications for analysts, admins, and leadership.
Section 1
AI4RA Introduction
This section introduces the AI4RA mission, operating principles, and ecosystem tools that connect workshop practice
to institutional research analytics needs.
Slide focus
AI4RA Objectives
Highlights three goals: open-source workflows, trustworthy AI tools, and partner-site implementation and impact assessment.
Slide focusCore Pillars for AI4RA
Defines accuracy, reproducibility, flexibility, and security as the foundation for responsible AI-enabled analytics.
Slide focusCommunity of Practice Ecosystem
Maps the ecosystem: Vandalizer, Data Lakehouse, AI Literacy Course, and additional AI4RA projects.
Slide focusBridge to Research Analytics
Frames AI4RA as a path to remove analysis barriers, address stakeholder expectations, and sustain governance practices.
Section 2
Data Science & AI Intersection
This section explains how AI changes both what teams can build and what stakeholders expect, then offers a practical
approach for implementation.
Slide focus
FAIR Principles as Guardrails
Uses Findable, Accessible, Interoperable, and Reusable principles as operational constraints for AI-supported data work.
Slide focusShift 1: Proactive Data Science
Notes faster system building, strong alignment with data science skills, and risk of overload from too many pathways.
Slide focusShift 2: Reactive Data Science
Describes growing demand for faster AI-enabled outputs, competitive pressure, and preference shifts toward chatbot interfaces.
Slide focusToken-by-Token Generation Exercise
Demonstrates that LLM outputs vary because generation is probability-based rather than fixed.
Slide focusSimplify the Approach
Presents a three-step flow: identify bottlenecks, classify automation/collaboration fit, and build a simpler solution.
Section 3
Sandbox Time
This section moves into hands-on practice in Vandalizer through three structured exercises focused on extraction and
cross-document checking.
Slide focus
Stepping Into the Vandalizer
Introduces the prompt workspace and emphasizes quick iteration for applied workshop tasks.
Slide focusExercise 1: Simple Extraction
Works with one NOFO document to build extraction terms, auto-generate terms, and export results.
Slide focusExercise 2: Repeated Extractions
Runs a pre-built NSF extraction process across multiple award notices and tests a known-missing term.
Slide focusExercise 3: Multi-Document Interactions
Compares budget documents, identifies inaccuracies, and checks alignment after adding research strategy context.
Section 4
Decision Frameworks
This section focuses on ownership, accountability, task fit, and interaction discipline for trustworthy AI use in
research administration contexts.
Slide focus
Who Owns AI-Generated Content?
Summarizes copyright and authorship concerns, including meaningful human contribution and institutional policy gaps.
Slide focusHuman ↔ AI Continuum
Positions work in three zones: human leads, human in the loop, and AI leads with validation checks.
Slide focusTask Classifier Activity
Invites participants to classify example analytics tasks along the continuum to surface decision criteria.
Slide focusThen and Now: Task Fit
Contrasts earlier DS-vs-AI distinctions with agentic workflows while preserving the principle that task fit matters.
Slide focusPrompt Engineering to Intent Engineering
Shifts emphasis from perfect prompts toward better processes while keeping prompts aligned with user intent.
Slide focusFour Disciplines for AI Interaction
Balances context-window use across intent, information, instructions, and conversation.
Slide focusAI Literacy Companions
Introduces companion roles (e.g., auditability, decomposition, intent, prompts, context, and evaluation support).