◇ Could this help me?
Yes. This is a prompt-only technique with no installs or third-party code, and its substance is genuinely good engineering discipline: it forces concrete eval metrics, failure-mode handling, and realistic scope instead of the vague toy apps AI tools default to, which is exactly what makes a portfolio credible. It is a reusable prompting pattern rather than a Claude Code configuration change, most useful when starting new build projects.
You are a principal engineer tasked with creating AI-native portfolio projects for the current job market. Generate 3 portfolio-ready projects with these constraints. No toy apps: each project must reflect real-world engineering concerns (backpressure, retries, idempotency, pagination, partial failures, schema migrations, observability). Use modern, industry-standard stacks: Next.js 14 (App Router); Node.js (Express or Fastify) or Python (FastAPI); TypeScript or Python; Postgres (Supabase), Redis, S3-compatible storage; Prisma or SQLAlchemy. Each project must be in a distinct domain (fintech, healthcare ops, marketplace tooling, infra, devtools) to show breadth. Avoid vague AI claims: define concrete eval metrics, latency targets, and cost ceilings. Scope must be realistic for a solo engineer or small team in 1 to 3 months. Keep explanations concise but technically dense. For each project, include a Claude Code pairing plan of 8 concrete prompts to paste during the build: 1 scaffolding (monorepo file structure), 2 contracts (OpenAPI and GraphQL schemas with typed clients), 3 data (Postgres schemas, migrations, seed scripts), 4 workers (queue plus one job with retries and dead-letter handling), 5 tests (Jest or pytest with example unit and integration tests), 6 AI (RAG or agent loops, evaluation harness, small golden dataset), 7 CI/CD (GitHub Actions for test, build, Docker push), 8 instrumentation (OpenTelemetry).