🔥 Grill Me — AI Architect Interview for Vibe Coders

"Senior-architect judgment, not generic AI guessing — grill me before you build."

The problem

You're vibe-coding your next side project. You describe an idea to an AI, get code back, and ship it. Three weeks later you discover: no auth model, the wrong database for your access pattern, an architecture that can't survive its first real traffic spike. Nobody asked the hard questions before the first line of code got written.

What Grill Me does

It runs the same relentless, one-question-at-a-time architecture interview a senior architect would put you through — except every recommended answer at a key decision point is grounded in 18 curated architecture reference files, not the model's generic training intuition.

Describe your idea
      │
      ▼
10-topic interview (one question at a time)
Product & Vision → Actors → Core Flows → Business Model →
Tech Stack → Service Topology → Integration Patterns →
Security & Compliance → Phase Plan & Risks → Infra & Observability
      │
      ▼
Locked decision log
      │
      ▼
Pick your docs: README · PRD · Architecture · Plan · Use-cases · Deliverables
      │
      ▼
Real files written to your own docs/ folder, one at a time
      │
      ▼
Closing skills-required audit of your own installed skills

Knowledge grounding — not just a claim

At the questions where it matters most, the recommended answer is traceable to a real reference file:

• Database choice → grounded in a 12-database-type decision tree (SQL, Document, Graph, Vector, and more — each with a "when to choose" checklist)
• REST auth method → grounded in a Basic/OAuth/API-Key/JWT decision guide
• Architecture style → grounded in a 10-style comparison (and it will tell you when Hexagonal is over-engineering for your project, not just push its own subject matter)
• Security posture → grounded in the OWASP Top 10, ranked by actual risk to your app's shape
• Delivery methodology → grounded in a 12-methodology comparison (Scrum vs Kanban vs Waterfall, matched to your team size)
• Deployment pattern → grounded in cloud design patterns + 12-factor app discipline
• Plus: NFRs, TOGAF, integration patterns, AI tooling stack, OO/GoF design principles, web design fundamentals, and general system design — 18 files total, each wired to a specific interview question.

Real example (from an actual dry-run)

Q: What database fits your data shape?
Recommended, grounded in database-tree.md: PostgreSQL — this is structured, relational data with straightforward joins between entities. The file's own decision tree flags Document DBs for a different case (flexible/hierarchical schema) that doesn't apply here.

Q: What architecture style fits?
Recommended, grounded in architecture-types.md + hexagonal-architecture.md: Modular Monolith, not full Hexagonal — the reference file's own "Poor Fit Scenarios" section lists small CRUD apps as a case where ports-and-adapters is unnecessary overhead. Notice it didn't just recommend its own subject matter.

What you get

• 7 SKILL.md files (the interview + 6 doc generators + a skills audit)
• 18 scrubbed, English-language reference files
• Every diagram (Mermaid sequence/state/topology) ships with a plain-English walkthrough underneath — no Mermaid literacy required to read your own architecture doc
• Zero code, zero runtime dependencies, zero external API calls — runs entirely inside your own local Claude Code

Stated plainly

• Knowledge base is a snapshot — it covers 18 areas well, not every architecture topic that exists
• No automatic update mechanism yet for future knowledge-base improvements
• Documentation-generation tool, not a substitute for a real architect on genuinely novel or heavily regulated systems

Install

1. Download after purchase
2. Copy the 7 skill folders into your project's .claude/skills/
3. Copy references/ into your project root
4. Describe what you're building — the interview starts automatically