🎓 Final Capstone Project (100 pts)

The task

Design — and partially prototype — a real system that exercises the core skills of this course. This is your chance to build something you can point to and say, "I designed and built this, and I can explain every decision in it." Your capstone must be your own project, chosen and scoped in Week 15, and it must integrate multiple layers of the stack you've learned across AIINFRA 100–303. A single Jupyter notebook that calls an API is not a capstone. A system with a deployment pipeline, a served or fine-tuned model, an agent or RAG layer, and a security/cost story is.

Requirements

Your submission must include all three of the following components:

1. A written design document (1,500–2,500 words, or equivalent in structured sections) that maps every major decision back to a specific skill or concept from Weeks 1–15. At minimum, it must address:
1. Problem & scope — what the system does, who it's for, and why it matters.
2. Architecture overview — the major components and how they communicate.
3. Containerization & deployment — how the system is packaged and what your CI/CD pipeline does (build, test, deploy stages).
4. Cloud & infrastructure — what's provisioned, where, and why (compute, storage, networking choices).
5. Model serving / fine-tuning — which model(s) you use, whether you fine-tuned or used a base model, and how inference is served (latency, throughput, scaling considerations).
6. Agents, MCP, and/or RAG — how the system reasons, retrieves, or takes action, and what tools or data sources it's connected to.
7. Security — your threat model, guardrails, access control, and data-handling decisions.
8. FinOps — estimated cost per request or per month at a stated usage level, and at least one cost-control decision you made deliberately.
9. Known limitations & what you'd do with more time.
2. A diagram of the system architecture — components, data flow, and trust boundaries. Any diagramming tool is fine (draw.io, Excalidraw, Mermaid, Lucidchart, even a clean hand-drawn diagram photographed clearly) as long as it's legible and included in or linked from your submission.
3. A working prototype artifact built with the course's free/low-cost toolchain (the same containers, cloud free-tiers, open-weight or low-cost hosted models, and open-source agent/RAG frameworks you've used all course). It does not need to be production-scale, but it must run and demonstrate the core interaction end-to-end — for example, a request that flows through your API, hits your served or fine-tuned model, uses a tool call or retrieval step, and returns a result.

Deliverables

• [ ] Design document (PDF or shared doc link), addressing all 9 sections above
• [ ] Architecture diagram (embedded in the design doc or attached separately)
• [ ] Link to your prototype's source repository (with a README covering setup and how to run it)
• [ ] Short (under 5 min) screen recording or set of screenshots showing the prototype actually running, in case a live grading run isn't possible
• [ ] A one-paragraph cost estimate summary, called out clearly in your design doc

Grading

Graded out of 100 points using the Capstone / Project Rubric. This assignment maps to all five course outcomes: