Hands-on projects¶
Fifteen guided builds: ten cloud, five AI. Each has an estimated time, a goal you can articulate to an interviewer, and inline code or commands. Most run on free-tier accounts; the AI builds list cheap cloud-GPU options where local hardware isn't enough.
Cloud builds (10)¶
| Build | Time | What you'll have at the end |
|---|---|---|
| Deploy a 3-tier app | 3-4 hours | LB → app servers → DB on AWS, Azure, or GCP |
| Build a CI/CD pipeline | 4-5 hours | Auto-build, test, and deploy on commit |
| Set up a monitoring stack | 3-4 hours | Prometheus + Grafana + alertmanager wired end-to-end |
| Implement zero-trust security | 4-5 hours | Identity-aware access, no-implicit-trust networking |
| Build a data pipeline | 4-5 hours | Ingest → transform → load with scheduled runs |
| Deploy an ML model | 3-4 hours | Trained model behind an API endpoint |
| Set up a Kubernetes cluster | 4-5 hours | Production-shaped cluster, ingress, observability |
| Build infra with Terraform | 3-4 hours | VPC + compute + DB declared as code, with state |
| Build a serverless app | 3-4 hours | API + queue + function + storage, event-driven |
| Run a DR drill | 4-6 hours | Tested failover, measured RTO/RPO |
AI builds (5)¶
| Build | Time | What you'll have at the end |
|---|---|---|
| Build a RAG pipeline | ~30 min | Docs → chunks → pgvector → Claude with retrieval + evals |
| Build a Claude agent with MCP | ~30 min | Agent SDK + custom MCP server reading files + SQLite |
| Run Llama on a single GPU | ~45 min | Open-weights model serving locally or on a cheap rented GPU |
| Set up an eval harness | ~30 min | Reproducible eval suite for prompt and model comparisons |
| Fine-tune with LoRA | 1-2 hours | LoRA-trained small model, merged + served + benchmarked |
How to pick¶
- First-ever cloud build: Deploy a 3-tier app. Touches networking, compute, identity, and storage in one project.
- First-ever AI build: Build a RAG pipeline. Most hireable AI pattern in 2026.
- Interview prep: Set up a Kubernetes cluster, Build infra with Terraform, Build a CI/CD pipeline. These map to the most common practitioner questions.
- Cheapest to run: the five AI builds (under $5 each on rented GPUs); serverless and Terraform on the cloud side (free-tier friendly).
- Most production-shaped: Implement zero-trust, Run a DR drill, Set up a monitoring stack.
Related¶
- Concepts - the "why" behind each pattern
- Architecture patterns - reference designs each build follows
- Topic indexes - all four pillars per subject
Which lab for which cert¶
Each lab maps to the certs it exercises. The reverse index lives in labs-by-cert.md.