Culminating integration course; students prove they can operate, secure, observe, document, and explain a complete AI infrastructure system.
← All courses
// AIINFRA 303 · Semester 4
Capstone Project
Design, build, deploy, and present one production-grade AI system
AIINFRA 303 is the culminating capstone of the AI Infrastructure and Architecture certificate sequence. Students design, build, deploy, and present a complete production-grade AI system of their own scoping — integrating compute, containers, cloud infrastructure, and a model or agent layer end to end. The course emphasizes professional practice: architecture documentation, infrastructure as code, CI/CD automation, security and observability, and a polished portfolio artifact suitable for employer review.
download / canvas package
Canvas IMSCC Export
Canvas IMSCC Export
Download the complete Canvas import package for AIINFRA 303.
01 / outcomes
Outcomes
Course objectives
- Scope, plan, and document a production-grade AI infrastructure project with clear requirements, success metrics, and architecture decision records
- Design and diagram a complete end-to-end AI system architecture integrating compute, containers, cloud, and the model/serving layer
- Implement, containerize, and deploy the system to the cloud with infrastructure-as-code and an automated CI/CD pipeline
- Integrate a model capability (serving, fine-tuning/adaptation, RAG, or an agent/MCP layer) with security guardrails, evaluation, and observability
- Load-test, right-size, document, and present a reproducible, cost-instrumented capstone with a professional portfolio artifact and demo
Student learning outcomes
- Scope, plan, and document a production-grade AI infrastructure project with clear requirements, success metrics, and architecture decision records.
- Design and diagram a complete end-to-end AI system architecture integrating compute, containers, cloud, and the model/serving layer.
- Implement, containerize, and deploy the system to the cloud with infrastructure-as-code and an automated CI/CD pipeline.
- Integrate a model capability (serving, fine-tuning/adaptation, RAG, or an agent/MCP layer) with security guardrails, evaluation, and observability.
- Load-test, right-size, document, and present a reproducible, cost-instrumented capstone with a professional portfolio artifact and demo.
02 / schedule
16-week schedule
Wk 01
Capstone Kickoff: Scoping a Production AI System
Kicks off the capstone by having students scope a production-grade AI system of their own choosing.
Wk 02
Requirements, Success Metrics, and Architecture Decision Records
Covers writing Architecture Decision Records and defining SLI/SLO success metrics for the capstone project.
Wk 03
System Architecture and Full-Stack Infrastructure Design
Covers layered AI system architecture and producing C4-style context and container diagrams as code.
Wk 04
Project Planning: Milestones, Risk, and a Reproducible Repository
Covers reproducibility practices, data versioning, a scaffolded repository, and a risk register with milestone roadmap.
Wk 05
Building the Core: Containerized Services and Environments
Students build out the containerized core services and environments defined in the Week 3 architecture sketch.
Wk 06
Deploying to the Cloud with Infrastructure as Code
Covers deploying the capstone's core services to the cloud using infrastructure as code.
Wk 07
A CI/CD Pipeline for Your Capstone
Covers CI/CD concepts and building a working GitHub Actions pipeline that lints, tests, builds, and pushes a Docker image.
Wk 08
Design Review and Working Vertical Slice
Midterm week: students deploy a working vertical slice to an orchestrator and present a design review against their ADRs, diagrams, and SLOs.
Midterm · covers Wks 1–7Wk 09
The Model Layer: Serving, Adaptation, or RAG Integration
Students choose and build the model-layer strategy for their capstone: serving, fine-tuning/adaptation, or RAG.
Wk 10
Adding Agentic and MCP or Retrieval Capabilities
Covers adding agentic, MCP, or retrieval capabilities on top of the capstone's model layer.
Wk 11
Security, Guardrails, and Responsible-AI Hardening
Covers hardening the capstone with security controls, guardrails, and responsible-AI practices.
Wk 12
Observability, Evaluation, and Cost Instrumentation
Covers instrumenting the capstone with Prometheus/Grafana observability, OpenCost cost allocation, and a RAGAS evaluation baseline.
Wk 13
Load Testing, Right-Sizing, and Performance Tuning
Covers load testing the capstone system and right-sizing and tuning it for performance.
Wk 14
Documentation, Runbooks, and Reproducibility
Covers writing documentation and runbooks that make the capstone reproducible for another operator.
Wk 15
Portfolio Packaging, Demo Prep, and Technical Storytelling
Covers turning the capstone repo into a portfolio artifact with an architecture-diagram README and a rehearsed demo script.
Wk 16
Final Capstone Submission, Presentation & Course Review
Final capstone week: students submit, present, and demo their completed production-grade AI system for course review.
Capstone03 / Senate-ready controls
Course evidence and boundaries
These controls make the curriculum reviewable without changing the 10-course sequence.
Complete production-grade AI system portfolio mapped to program learning outcomes and reviewed against the shared rubric.
Full pipeline/security evidence, including scan results, mitigations, access controls, governance note, and incident response path.
Public or reviewable repo, CI/CD logs, deployment notes, security scan report, architecture diagram, cost estimate, accessibility note, runbook, and recorded demo.
Capstones may specialize in AWS, Azure, or GCP, but every required outcome must have a local/no-cost fallback.
04 / foundry projects
Foundry-backed assignments
These assignments are written in platform-agnostic language. They assess infrastructure evidence, not familiarity with a single vendor console.
AI Foundry Fit Check
Every capstone proposal states whether shared compute is required, what profile is requested, what services will persist, and what runtime evidence will be collected.
Evidence: Foundry yes/no decision, workload type, compute profile, data classification, access pattern, expected logs/metrics, and fallback plan.
STEM Research Infrastructure Studio
CS/CIS students partner with STEM students or faculty: STEM defines the research question and validation; AIINFRA teams build the reproducible compute, data, benchmark, and monitoring environment.
Evidence: Joint charter, data card, baseline benchmark, reproducibility package, research-facing summary, and technical appendix.
Foundry Evidence Appendix
Foundry-backed capstones document what ran on shared infrastructure and why the selected compute profile was appropriate.
Evidence: Deployment evidence, operational metrics, security/governance notes, cost/right-sizing memo, model or adapter references, and runbook.
05 / tools
Tools & frameworks
Planning & Architecture
Containers, Cloud & IaC
CI/CD
Model Layer
Security & Governance
Observability, FinOps & Portfolio