From your first ghost-text suggestion to a deployed, AI-reviewed service. Twelve hands-on modules covering every feature: completion, chat, context files, agent mode, MCP, Gemini CLI, GitHub PR reviews, Google Cloud, and enterprise controls.
From G02 onward every lab builds the same small Flask API — you generate its storage layer from comments, debug it in chat, test it, extend it with agents, get it reviewed on GitHub, and deploy it to Cloud Run. The capstone then has you build a second service, Meal Planner, entirely on your own using every skill.
The feature map, the three editions and pricing, the June 2026 Antigravity transition, and how the 12 modules fit together. No code.
G01 · SETUPInstall in VS Code and JetBrains, sign in, trigger and verify your first AI-generated function. Plus the Standard/Enterprise admin flow.
G02 · COMPLETIONHow completion really works ("your file is the prompt"), the spec-comment formula, partial accepts — and Recipe Box's storage layer built from comments.
The code-aware chat panel, /fix /explain /doc /simplify /generate, right-click smart actions, source citations — and a 3-bug debugging lab.
G04 · QUALITYNatural-language code transformations, unit test generation, and why failing generated tests are good news (your design just got reviewed).
G05 · CONTEXTWhole-codebase awareness, standing rules that visibly change output (proven by experiment), and mechanical secret-exclusion.
The plan → diff → approve loop, checkpoints, command gates — and a coordinated 5-file tagging feature shipped through them.
G07 · AGENT IIConnect agents to GitHub via MCP, restrict built-in tools, the honest truth about auto-approve — and an agent that fixes a real GitHub issue.
G08 · TERMINALThe same brain in your shell: @files, !commands, headless mode as a Unix tool — generating READMEs and changelogs from pipes.
The PR-review app: summaries, severity-labeled findings, /gemini commands, config.yaml and styleguide.md — tuned on your own repo.
G10 · CLOUDDeploy Recipe Box to Cloud Run from chat (/deploy), plus the Gemini surfaces in BigQuery, Firebase, and beyond. Includes a no-cloud fallback.
G11 · CAPSTONECode customization, data governance, IP indemnification — then build, test, review, and deploy the Meal Planner service using everything, graded against a 7-point rubric.
Four standalone capstones anchored in real domains — each runs the complete Gemini Code Assist workflow (rules-first GEMINI.md, comment-driven generation, generated tests, an agent-mode feature, AI PR review) against problems with genuine industry stakes. Do the one matching your career track, or all four for a full-stack portfolio.
Public-records lien filings → lender-ready risk data. Medallion architecture (bronze/silver/gold), entity resolution, amendment state tracking, quarantine-based data quality.
CAPSTONE · APIPurchase-order lifecycle with an enforced, audited state machine; pluggable carrier adapters; business-day SLA breach detection. Exhaustive 64-pair transition testing.
CAPSTONE · BACKENDA policy-as-data clinical rules engine: CPT/ICD criteria in versioned JSON, explainable decisions with reason codes, immutable audit trails, PHI-safe logging discipline.
CAPSTONE · FRONTENDA zero-build vanilla JS operations dashboard: live order board, SLA banner, accessible detail drawer, honest loading/error/empty states, headlessly tested logic, dual themes.