# How the AI-Native Claude-Certification Tutor Works: Architecture and Execution Flow > Discover the AI-Native Claude-Certification Tutor. Learn how its skill-based architecture and stateful tracking transform AI Engineering from Scratch into an interactive learning coach for Claude certifications. - Repository: [Rohit Ghumare/ai-engineering-from-scratch](https://github.com/rohitg00/ai-engineering-from-scratch) - Tags: architecture - Published: 2026-09-01 --- **The AI-Native Claude-Certification Tutor transforms the AI Engineering from Scratch repository into an interactive learning coach for Anthropic's four Claude certification tracks, using a skill-based architecture defined in [`skills/claude-certification/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/claude-certification/SKILL.md) and stateful progress tracking via [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md).** The rohitg00/ai-engineering-from-scratch repository implements a self-contained, **AI-Native Claude-Certification Tutor** that guides learners through the CCAO-F, CCDV-F, CCAR-F, and CCAR-P exam tracks without requiring external services. Unlike traditional study guides, this system operates as an executable workflow that validates hands-on labs, administers official assessments, and persists learner state directly in the repository. The tutor's behavior is governed by structured JSON definitions and a comprehensive skill specification that orchestrates four distinct operating modes. ## Core Components and File Structure The tutor relies on a hierarchy of declarative files that separate pedagogy from execution logic. ### The Skill Definition and Program Manifest The **skill definition** in [`skills/claude-certification/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/claude-certification/SKILL.md) serves as the tutor's operational brain. This file defines the four operating modes—**Onboarding**, **Lesson**, **Assessment**, and **Remediation**—and specifies the step-wise interaction protocol for each. Complementing this, the **program manifest** at [`certifications/claude/program.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/certifications/claude/program.json) provides canonical metadata including the provider, version, and verification date, while enumerating the four available certification tracks. ### Track Definitions and Lesson Assets Each certification track maintains a dedicated JSON configuration under `certifications/claude/tracks/`. For example, [`certifications/claude/tracks/ccao-f.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/certifications/claude/tracks/ccao-f.json) declares the ordered lesson list, domain weight distributions, assessment mappings, and study plans specific to the Claude Certification Associate Operator - Foundational exam. Individual lessons follow a strict directory convention under `certifications/claude/lessons//`: - [`docs/en.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/docs/en.md) – Human-readable explanation of concepts - [`code/main.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/code/main.py) – Runnable validator or scenario runner - `code/tests/` – Unit tests that must pass for lesson completion - `outputs/` – Reference artifacts (e.g., markdown contracts) for validation - [`quiz.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/quiz.json) – Pre-check and post-check assessment questions ### Assessment Definitions and State Management Diagnostic and mock exams reside in `certifications/claude/assessments//` as static JSON files. The tutor imports these verbatim and never generates new questions, ensuring alignment with official exam objectives. Learner progress persists in [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md) at the repository root, which the tutor updates atomically after each interaction to preserve route order, domain readiness scores, and assessment history. ## The Four Operating Modes The AI-Native Claude-Certification Tutor selects its operating mode based on the presence and content of [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md), creating a stateless yet state-aware execution model. ### Onboarding Mode When no state file exists, the tutor initiates **Onboarding Mode** to capture learner context. It presents the independence disclaimer, then asks three critical questions: desired outcome, prior experience level, and weekly study hours. Based on these inputs, the system maps the learner to an appropriate track and generates a starter [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md) containing the goal, active track code, start date, and an initialized route table. ### Lesson Mode In **Lesson Mode**, the tutor orchestrates a structured learning loop for each unfinished lesson. It first recalls the previous lesson's quiz results, then sequentially explains the problem statement, underlying concepts, and interactive lab requirements. For technical validation, the tutor executes the lesson's validator and unit tests: ```bash python3 certifications/claude/lessons/05-output-evaluation-and-validation/code/main.py python3 -m unittest discover -s certifications/claude/lessons/05-output-evaluation-and-validation/code/tests -v ``` For CCAO-F learners preferring conceptual study, the tutor offers a **guided-no-code path** that executes [`main.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/main.py) on behalf of the learner and interprets results in plain language. The system validates generated artifacts against reference outputs in the lesson's `outputs/` directory. Completion requires both a post-quiz score of at least 70% and passing lab validation, at which point the tutor updates the route table to "Complete". ### Assessment Mode **Assessment Mode** loads exact assessment definitions from the track's JSON configuration and presents questions one-by-one. Scoring relies on exact answer set matching rather than semantic similarity, ensuring objective evaluation. Upon completion, weak domains are automatically added to a review queue within [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md), and results are appended to the persistent state file. ### Remediation Mode Following diagnostic or mock assessments, **Remediation Mode** analyzes domain performance and highlights knowledge gaps. The tutor cross-references weak domains against the track's `deepDives` list to suggest targeted lessons, then re-opens specific sections of the learning route for focused study. ## Execution Flow and Validation Logic The tutor's execution follows a strict protocol defined in the skill file. After mode selection, the system enforces the "explain-build-verify" loop through code execution and artifact validation. Reference outputs in `certifications/claude/lessons//outputs/` serve as immutable rubrics; learners generate their own artifacts under `learning-artifacts/claude///` for comparison. This **source-of-truth-first** architecture ensures all routing decisions derive from [`program.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/program.json) and track definitions rather than model hallucinations. ## Running the Tutor: Practical Commands While the tutor orchestrates these commands automatically, learners can execute validation steps manually: ```bash # Start a diagnostic assessment for the CCAO-F track python3 -m tutor claude-certification --mode assessment --track ccao-f --assessment diagnostic # Validate a specific lesson's code and tests python3 certifications/claude/lessons/03-prompt-engineering-basics/code/main.py python3 -m unittest discover -s certifications/claude/lessons/03-prompt-engineering-basics/code/tests -v ``` ## Summary - The **AI-Native Claude-Certification Tutor** operates through four modes: Onboarding, Lesson, Assessment, and Remediation, defined in [`skills/claude-certification/SKILL.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/skills/claude-certification/SKILL.md). - **State persistence** occurs in [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md), enabling stateless invocation with automatic context recovery. - **Lesson validation** combines executable Python validators ([`main.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/main.py)), unit tests, and artifact comparison against reference outputs. - **Assessment integrity** is maintained by using verbatim JSON exam definitions from `certifications/claude/assessments/` with exact-match scoring. - **No-code support** for the CCAO-F track allows non-technical learners to complete labs through tutor-mediated execution. ## Frequently Asked Questions ### How does the tutor track progress without a database? The tutor implements **file-based state management** through [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md) at the repository root. This markdown file contains the learner's route table, domain readiness scores, and assessment history. After each lesson or exam, the tutor atomically updates this file, treating the filesystem as the persistent store. This design eliminates external dependencies while maintaining continuity across sessions. ### Can I complete the CCAO-F track without writing Python code? Yes. The tutor provides a **guided-no-code path** specifically for the CCAO-F foundational track. In this mode, the tutor executes the lesson's [`code/main.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/code/main.py) and unit tests on behalf of the learner, then explains the results and validation outcomes in plain language. This preserves the "explain-build-verify" pedagogical loop while accommodating learners who focus on operational concepts rather than implementation details. ### Where are the exam questions stored? All assessment content resides in static JSON files under `certifications/claude/assessments//`. The tutor loads these definitions verbatim from the track configuration files (such as [`ccao-f.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/ccao-f.json)) and never generates new questions via the model. This ensures that diagnostic and mock exams remain consistent with the official certification objectives and do not drift due to AI hallucination. ### What triggers a lesson to mark as complete? A lesson achieves "Complete" status only when two conditions converge: the learner scores at least 70% on the post-lesson quiz defined in [`quiz.json`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/quiz.json), and all unit tests in `certifications/claude/lessons//code/tests/` pass successfully. The tutor validates generated artifacts against reference outputs in the lesson's `outputs/` directory before updating the route table in [`CLAUDE-CERTIFICATION.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/CLAUDE-CERTIFICATION.md).