🎓 Claude Code Architect Certification Tutor (CCAF Prep)
Claude Code Architect Certification Tutor (CCAF Prep) is a production-grade, AI-powered learning platform designed for software architects mastering Claude Code and agentic systems. Built on TanStack Start (React 19) and Supabase PostgreSQL with pgvector, the application features an interactive CodeCanvas environment with live in-browser execution (Pyodide WASM for Python and isolated Web Workers for JavaScript), a multi-agent AI mentor suite providing real-time voice streaming and line-by-line architectural breakdown matrices, and personalized FSRS spaced-repetition study plans powered by a candidate onboarding wizard. The system is engineered for enterprise security and scale, featuring tier-aware model routing, an encrypted server-side BYOK key vault (AES-256-GCM), automated daily rate-limiting guards, mobile-responsive accessible drawers (WCAG/ARIA), resilient route loading boundaries, router-level SEO with Open Graph cards, and hardened Supabase RLS security policies.
🌟 Key Feature Highlights
💻 1. CodeCanvas Live Execution Environment
- Multi-File Syntax Highlighted Tabs: Read-only and editable multi-file tabs for Python and JavaScript codebases.
- In-Browser Execution Engine:
- Python: Client-side WASM execution via Pyodide with instant standard library support.
- JavaScript: Isolated Web Worker sandbox execution via Blob Workers.
- Execution Controls: 10-second timeout enforcement, execution cancellation, live streaming console output (stdout/stderr/return values).
- Diagnostic Tracebacks: Line-number error highlighting, stack trace displays, and pre-run syntax validation hooks.
- Telemetry Logging: Owner-scoped
code_executions database logging tracking latency, success rates, and payload metrics.
🤖 2. 5-Agent Autonomous Code Generation Loop
- Multi-Agent Orchestration Pipeline:
- Research Agent: Grounded library gap brief synthesis.
- SME (Subject Matter Expert) Agent: Code example generation.
- Verifier Agent: Static analysis & sandbox constraint verification.
- Documentation Agent: Line-by-line code walk generation.
- Advice Generator Agent: Multi-dimensional advice matrices (walks, design tradeoffs, misconception checks, concept linkages, follow-up questions).
- Feedback & Retry Loop: SME → Verifier loop retried up to 3 times with feedback; broken attempts discarded before saving to
codebases.
- Admin Live Stream Tracker: Real-time multi-agent generation tracker in section 12 of the Admin Console.
🔐 3. Encrypted Server-Side BYOK Key Vault
- AES-256-GCM Vault: Encrypted storage for candidate operational Anthropic (Claude) and Google (Gemini) API keys.
- Provider Verification: Live API key verification and testing before saving.
- Proxy Override Gates: Active unpaused vault keys securely override proxy subscription balances, routing inference directly through the candidate's custom operational key.
⚡ 4. Model Routing, Usage Analytics & Rate Limits
- Cache-First Model Routing: Instant resolution via
ai_response_cache and tier-aware model allocation (profiles.membership_tier).
- AI Usage Analytics: Detailed logging of token spend, cache hit rates, cost savings curves, and model breakdown charts in the Admin Usage Board.
- Rate Constraint Guards: Tier-based daily request/token quotas and burst throttles across mentor streams, codegen actions, and TTS calls (
rate_events), with BYOK user exemptions.
🧠 5. Adaptive Learning Engine & Onboarding
- Interactive Onboarding Wizard: First-time candidate flow capturing target exam date, desired score band, weekly study hours, and self-assessed domain confidence.
- FSRS Spaced Repetition: Automated study plan seeding, review scheduling, and diagnostic pass probability forecasting bands.
- Study Canvas Tabs: Context-bound Code, Video, and Documentation sections tied to FSRS retention metrics.
📊 6. Admin Intelligence & Content Control Desk
- Admin Usage Board (Section 16): Cache hit ratios, credits spent vs. saved, model token volume trends.
- Spider Control Desk (Section 13): Catalogued source URL crawler with configurable crawl intervals and
Crawl_Due batch triggers.
- Corpus Defragmentation Sweep (Section 14): Automated section consolidation, empty artifact stripping, and chunk re-embedding.
- Coverage Parity Matrix (Section 15): Bank composition vs. exam blueprint weights, difficulty mix, and citation tracking.
여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗
