Wirewalk Instituteof AI
edu-220 Intermediate AvailableFlagship

Building and Deploying AI in Higher Education and Research Computing

For the university IT, research computing, research compliance and sponsored programs staff who build and run AI services, taught through the specific jobs they do: scoring assistant contracts, writing Slurm GPU partitions, running a campus LLM gateway, closing out a dbGaP enclave, and setting a GPU rate.

Level
Intermediate
Study time
15 hours
Lessons
19 in 6 modules
Exam
20 questions

About this course

This course is organized around the specific tasks university staff perform when they put AI into service, not around general themes. Each module follows one or two named roles through a concrete job: what the task looks like today, where AI helps and where it must not be used, the data and systems actually available, the products used in higher education as of October 2026 (named for teaching, never endorsed), the rules that apply to that exact task, how to measure success, and a hands-on exercise built on synthetic data. No real student, patient, participant or proposal data is used anywhere.

You will score fictional assistant term sheets against the FERPA direct-control test and the GLBA Safeguards Rule's service-provider requirements. You will trim the scopes an AI tutor requests through LTI 1.3 and write integrity guidance consistent with Turnitin's statement that its AI score should not be the sole basis for adverse action. You will size whole-GPU, MIG and shard partitions from a synthetic sacct export and write Slurm 26.05 gres.conf, partition, QOS and billing settings. You will configure an OpenAI-compatible gateway over vLLM with SSO, per-team budgets and data-class routing, then audit its logs for planted faults. You will scope an NIST SP 800-171 enclave boundary on a shared cluster, classify AI workflows on controlled-access genomic data under NIH NOT-OD-24-157 and NOT-OD-25-081, and write the closeout checklist that removes model parameters. You will specify an on-premises Whisper transcription workflow for identifiable interviews and measure its word error rate. You will resolve deemed-export vignettes, route DoD contracts carrying DFARS 252.204-7012 and CUI to the 800-171 environment, and enforce a technology control plan in Slurm that also covers privileged administrators. You will apply NIH NOT-OD-25-132 (including its six-application cap, counted by Advisory Council year), NIH NOT-OD-23-149 and NSF's merit-review notice to proposed AI uses, test an AI pre-award checker against seeded errors, justify GPU-hours for ACCESS, the NAIRR or DOE allocations, set a GPU rate under 2 CFR 200.468, and evaluate a Tier-0 support virtual agent. Program, product and policy status is stated as of October 2026. Regulatory content is educational, not legal advice.

What you will be able to do

  • Score campus AI assistant contract terms against a FERPA direct-control (34 CFR 99.31(a)(1)(i)(B)), GLBA 16 CFR 314.4(f) and data-classification checklist, and specify the contract, retention and connector changes required before launch
  • Remove unneeded LTI 1.3 scopes from an AI tutor registration and write instructor guidance for a flagged submission that does not rely on a detector score alone
  • Design a GPU partition layout (whole GPU, MIG, shard) from Slurm accounting data and write Slurm 26.05 gres.conf, partition, QOS and TRESBillingWeights settings that keep regulated work off unisolated shards
  • Configure an OpenAI-compatible campus LLM gateway over vLLM with SSO, per-team budgets, data-class routing and a log retention limit, and find the policy violations planted in its request log
  • Scope an 800-171 enclave boundary on a shared cluster, classify AI workflows on controlled-access genomic data as allowed, needing NIH approval or prohibited under NOT-OD-24-157 and NOT-OD-25-081, and write a closeout checklist that removes model parameters
  • Specify an IRB-ready on-premises transcription and AI-assisted coding workflow for identifiable interviews, including consent language naming each processor and a word-error-rate and hallucination check
  • Classify proposed AI uses by PIs, students and reviewers under NIH NOT-OD-25-132, NIH NOT-OD-23-149 and the NSF merit-review notice, and resolve deemed-export questions for cluster and model access with a Slurm-enforced technology control plan, routing DFARS 252.204-7012/CUI work to the 800-171 environment
  • Compute a GPU-hour request with contingency for an ACCESS, NAIRR or DOE allocation, and a break-even GPU service rate that is identical for federal and non-federal users under 2 CFR 200.468

Syllabus

6 modules · 19 lessons · about 15 hours

Module 1: The IAM Lead and LMS Admin: Rolling Out Campus AI Assistants and an AI Tutor

An enterprise architect scores assistant contracts and wires identity, retention and connectors, and an LMS administrator admits an AI tutor through LTI 1.3 with minimal scopes.
3 lessons
  1. Scoring Assistant Term Sheets: ChatGPT Edu, Gemini for Education, Claude for Education and Microsoft 365 Copilot Against FERPA and GLBA
  2. Wiring SSO, SCIM, Retention and Connectors Without Surfacing Overshared Files
  3. Admitting an AI Tutor Through LTI 1.3: Trimming Scopes and Writing Integrity Guidance That Does Not Rest on a Detector Score

Module 2: The Research-Computing Engineer: Adding GPU Partitions in Slurm and Open OnDemand

An HPC engineer turns 30 days of accounting data into a whole-GPU, MIG and shard layout, writes the Slurm configuration, and stops interactive sessions from stranding GPUs.
3 lessons
  1. Reading a Month of sacct: Sizing Whole-GPU, MIG and Shard Partitions for a Class, Notebook Users and a Fine-Tuning Lab
  2. Writing gres.conf, Partition Stanzas, QOS Caps and TRESBillingWeights in Slurm 26.05
  3. Open OnDemand Jupyter and RStudio Sessions That Do Not Strand Idle GPUs

Module 3: The Research Software Engineer: Running a Campus LLM Gateway

An RSE team serves open-weight models with vLLM behind an authenticated gateway with virtual keys, budgets, data-class routing and a log retention limit, then audits it.
3 lessons
  1. Serving Open-Weight Models With vLLM: Capacity, Model Licenses and What Must Never Reach the Service
  2. Gateway Configuration: SSO, Virtual Keys, Team Budgets and Routing Restricted Data Only to On-Premises Models
  3. Auditing a Week of Gateway Logs: Shared Keys, Misrouted Personal Data, Budget Overruns and Retention Breaches

Module 4: The Data Steward, IRB Analyst and Export-Control Officer: Regulated Research Data on Shared AI Infrastructure

Three compliance roles decide what AI work may run where: an 800-171 enclave for dbGaP data, on-premises transcription of identifiable interviews, and foreign-national access to GPUs and model weights.
4 lessons
  1. Building an NIST SP 800-171 Enclave for AI on dbGaP Data Under NOT-OD-24-157
  2. Closeout Under NOT-OD-25-081: Classifying Enclave Workflows and Removing Model Parameters as Data Derivatives
  3. On-Premises Whisper and AI-Assisted Coding for Identifiable Interviews: Protocol, Consent and a Word-Error-Rate Check
  4. Deemed Exports on the Cluster: Screening, the Fundamental Research Exclusion and a Technology Control Plan Enforced in Slurm

Module 5: The Sponsored-Programs Officer: AI Rules for Proposals, Peer Review and Pre-Award Checks

An Office of Sponsored Programs turns NIH and NSF AI notices into campus procedure and tests an AI pre-award checker against seeded errors before trusting it.
3 lessons
  1. Turning NIH NOT-OD-25-132 Into Campus Procedure: Substantially AI-Developed Applications and the Six-Application Cap
  2. Keeping Reviewers Out of Chatbots: NIH NOT-OD-23-149 and the NSF Merit-Review Notice
  3. Piloting an AI Pre-Award Checker: Measuring Recall and Precision on Seeded Proposal Errors

Module 6: The Research-Computing Director: Allocations, GPU Rates and the Support Desk

A facilitator justifies GPU-hours for national allocations, the director sets a non-discriminatory GPU rate and demand forecast, and the IT service-management lead evaluates a Tier-0 virtual agent.
3 lessons
  1. Choosing ACCESS, the NAIRR or DOE ALCC and INCITE, and Justifying GPU-Hours From Measured Scaling Runs
  2. Setting a GPU-Hour Rate Under 2 CFR 200.468 and Forecasting Next Year's Demand in Three Scenarios
  3. A Tier-0 Virtual Agent for IT and Cluster Support: Identity Proofing, FERPA in Tickets and Stale-Article Evaluation

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by OpenAI, Google, Anthropic, Microsoft, Instructure, Turnitin, SchedMD, NVIDIA, Ohio Supercomputer Center (Open OnDemand), vLLM, BerriAI (LiteLLM), ATLAS.ti, MAXQDA, Descartes Systems Group, ServiceNow, or TeamDynamix. Product and company names are trademarks of their respective owners.

More in AI in Education and Research

All 9 courses in this track
edu-001 Free intro Available

AI in Education and Research: The One-Hour Briefing

A free hour built around seven real jobs in schools, campuses, labs and training teams: leveling a reading passage, drafting rubric feedback, handling an AI-detector flag, screening abstracts, writing and reviewing an NIH application, vetting a classroom AI tool, and turning an SOP into a microlearning module. Each shows where AI helps, where it must not decide, and the rule that applies.

1 h · 7 lessons Free intro
edu-110 Basic Available

AI for Special Education and Student Support Professionals

Work through a fictional student-support case using supplied instructional materials, observations and draft records. Practice useful AI drafting while preserving student participation, data boundaries, evaluation authority and measurable evidence.

12 h · 18 lessons See price at checkout
edu-210 Intermediate Available

Building and Deploying AI in K-12 Schools

Evaluate a California district’s AI deployment through privacy intake, tested access and minimum data, a cautious tutor evaluation, funded safety response, subgroup model/scoring audits and usable individualized communications. Build a cumulative board dossier using synthetic materials, six module Decision Labs and a strategic capstone.

24 h · 22 lessons See price at checkout