2027 Claude Code: Software Engineering, SDLC & CI/CD

This course contains the use of artificial intelligence. This course is software engineering with Claude Code, not a shallow walkthrough. You take one real system through the software development life cycle (SDLC), the way it happens in industry: there is a legacy system, and you need to upgrade it. So you can correlate every tip you see here directly with your own work. And you do it the way it is done in industry nowadays: step by step, in a structured manner, instead of an unstructured, unorganized one.

The project. You join Orion as its new engineer. To be clear, Orion is a fictional company made for this course, and its customers, tickets and messages are synthetic data. The software is real. Orion Classic, the old support desk, is a working Flask app, live on Vercel, with a TiDB Cloud database behind it. Your job across the whole course is to study it and replace it with a better desk, built with Claude. Project lessons open with where you stand, what you are asked to do and why it matters to the people at Orion.

The path: understand, plan, build.

• Understand: watch how the old desk is used, restore your own lab copy, and investigate the site and its database with Claude, checking each finding against the source.

• Plan: turn the evidence into a backlog, a decision record, requirements, a cost sheet, a proposal and a reviewed design document.

• Build: migrate the data, then build the API, the dashboard, search and reports, and ship through pull requests, checks and preview deployments.

What you build

• A replacement support desk: a FastAPI back end, a Next.js front end and a PostgreSQL-compatible TiDB database with the migrated Orion data.

• Semantic search with embeddings, and SQL reports you verify.

• A CI/CD route: pull request, checks, preview, merge, live check.

• A grounded assistant that answers from Orion's records, and a bounded triage agent that proposes and never decides alone.

• Gmail support over IMAP and SMTP: complaints synced to tickets, replies drafted, a person approves, the reply is sent.

What you learn, section by section

1. Welcome: the project walkthrough, the course resources (cheat sheets, the code and section branches) and installing your tools on a Mac or Windows.

2. Claude orientation: Chat, Design, Code and the API, plans and usage limits, models and effort, settings, and which Claude Code surface to use.

3. Join Orion: the legacy system, the people who use it, a checked backlog, a TiDB Cloud database, your lab, Auto Mode, a first MCP server, an architecture review and a replacement plan.

4. Claude fundamentals: projects, the composer, prompting, memory, voice, /rename, /init, /memory, project memory files, the ~/.claude folders, /add-dir, Plan Mode, /plan, steering, /rewind, /resume, /branch, /fork, /compact, /model, /effort, /sandbox, /permissions and Claude Design.

5. Investigating with Claude: web search with citations, Deep Research, connectors, a read-only database connection and charts in chat.

6. Plans and documents: an agentic modernisation plan, requirements, a proposal deck and a Google Sheets cost model.

7. Designs and artifacts: a design system, a replacement queue, a design document, a KPI dashboard, stateful artifacts, versions and sharing.

8. Reusable workflows: Agent Skills, a custom skill, plugins, scheduled tasks, Claude in Chrome and connected email drafts.

9. Building the replacement: data migration, FastAPI, the dashboard and case workspace, semantic search with /btw, SQL analysis and CI/CD.

10. Claude Design: critique, redesign from a reference and real data, build the design with Claude Code, and deploy.

11. Power tools: /code-review, /security-review, /simplify, /verify, /skills, subagents, /batch, /hooks, /mcp, /plugin, /output-style, /statusline, /cost and /design.

12. Parallel agents and automation: worktrees, /background, agent teams, /workflows, /goal, /teleport, /remote-control, headless review with claude -p in GitHub Actions, /schedule, /chrome, /insights, /status and managed settings.

13. API, SDK and gateways: the Anthropic SDK, tool calling, server tools, structured outputs, prompt caching, batch, files, the Claude Agent SDK, gateways and provider adapters.

14. Support communication: Gmail, investigation across mail and SQL, approved replies, and a final capstone on the live desk.

15. Optional, local models on Windows: Ollama and a comparison with a hosted model. This section is optional and Windows only. You can skip it and still finish the project.

What is run and what is only shown. Section 13, the Claude API section, runs through OpenRouter, starting on a free model, with a low-cost paid model on small samples where the free one is not enough. Anthropic and OpenAI are shown as provider settings, not run. Hosted agent runtimes are shown, not run. Managed settings are explained, not installed.

Real, not staged. You see the real Claude app, Claude Code, GitHub, Vercel and TiDB Cloud. When something breaks on camera, you see it and you see the fix.

Mac and Windows. The course is recorded on a Mac. Windows lessons cover the setup steps that differ.

Cheat sheets. Sections 2 to 15 each come with a PDF cheat sheet of their commands, prompts and shortcuts, attached to the section's first lecture. Longer sections have theirs in parts.

Also covered along the way: prompt engineering, Claude AI in chat and on the desktop, Claude Design, Claude skills, MCP servers, subagents and AI agents, agentic workflows, code review, GitHub Actions, the Claude API and the Claude Agent SDK.

Captions. Every lecture comes with English captions (WebVTT).

Who it's for: working software professionals who know programming, the terminal, Git, basic SQL and web apps, and who want to use Claude and Claude Code properly on the kind of work they already do.