₹1,9993 months
200+ students benefited
Claude Certified Architect: agents, Claude Code and MCP

Pass the Claude Certified Architect exam. In 10 sessions.

All five exam domains, two sessions each. A lab in every session, 120 scenario questions and two timed mock exams.

  • All 5 exam domains
  • 120 scenario questions
  • 2 timed mock exams
Built by Samir KumarEnterprise Architect · 21+ years in AI
TrendingYour support agent stops after its first tool call and returns a half answer. The loop checks whether the reply text contains "final answer". What is the right fix?Claude replies with two tool_use blocks in one message: look up the order and check the policy. How should your code send the results back?The user told the coordinator to cover only 2024 to 2026. A subagent's findings are from 2019. What is the most likely cause?A coordinator built with the Claude Agent SDK never delegates; it tries to do all the research itself. Its subagents are defined correctly. What should you check first?A senior developer put the team's testing conventions in ~/.claude/CLAUDE.md. A new teammate's Claude Code does not follow them. What is the cause?Your test files live next to the code in dozens of folders. You want one set of testing conventions applied to every test file, and nowhere else. What is the best fit?A developer must replace the logging library across 45 files, and some modules use it in unusual ways. How should they start in Claude Code?A stack trace points to one missing null check in one function. Which approach is best?Your automated pull request reviewer is told to "be conservative". Results still vary from file to file. What is the most effective change?After a month, developers ignore every review comment, although the security findings are almost always correct. Measurement shows style comments are wrong 70% of the time. What should you do first?▸
New course · open for enrolment
10 sessions10 videos10 labs114 lab tasks120 scenario questions2 mock exams
What you will practise

The decisions the exam asks you to make, built by hand.

An agent loop that reads stop_reasonAgentic loops

An agent loop that reads stop_reason

Run the tools Claude asks for, return each result with its ID, and stop on end_turn, not on a guess.

Session 1 ▸
Rules an agent cannot skipOrchestration · Hooks

Rules an agent cannot skip

Coordinators and subagents that get their context on purpose, and hooks that block what must never happen.

Session 2 ▸
Plan first, then change many filesClaude Code · Workflows

Plan first, then change many files

Plan mode, long sessions, and a CI review that reports only what matters.

Session 4 ▸
Extraction you can trustStructured output

Extraction you can trust

A tool schema, nullable fields, meaning checks, retries with the real error, and batches for the bulk.

Session 6 ▸
One team, the same servers, no secrets in gitMCP · Integration

One team, the same servers, no secrets in git

Project-scope MCP config with environment variables, resources and clear tool names.

Session 8 ▸
Escalate for the right reasonsReliability

Escalate for the right reasons

Explicit requests, policy gaps and no progress, not sentiment. Errors that carry partial results.

Session 10 ▸
See it for yourself

Two short videos, before you decide

First how it works, feature by feature. Then a tour of all four courses and which one is for you.

How it works, in one video

Every feature on the real screens: learn in the virtual class, build it in the lab, explain it with visual interview answers, then find jobs, tailor your resume and practise with a virtual interviewer that scores every answer.

  • Virtual class
  • Hands-on labs
  • Visual interview answers
  • Job search with a fit score
  • Tailored resume
  • Virtual interviewer and scored report
2 min 28 s

A tour of all four courses

The four courses with their numbers, which one is for you, and how every course works on real screens: the lesson, the lab, the practice questions, then the AI interview coach that finds jobs, tailors your resume, interviews you out loud and scores every answer.

  • Forward Deployed Engineer
  • Claude Certified Architect
  • AI Agent Developer
  • AI Engineer
2 min 48 s
All included

One course. Everything you need to walk in ready.

Each piece works on its own, and even better together. All of it opens the moment you pay.

10 sessions
samirgenaiacademy.com
The course page: the session list on the left and the narrated video

10 sessions, 10 narrated videos

10 sessions across the five exam domains

Every session has a narrated video you can follow step by step, a lab where you write the code, and 12 scenario questions with the answer and why.

See the syllabus
Practice labs
samirgenaiacademy.com
The lab room: the task on the left, an editor and a terminal on the right

10 labs, 114 tasks

Easy, Moderate and Advanced · runs in your browser

The lab opens full screen: the task on the left, an editor and a terminal on the right. You write the missing lines, run the code and press Check. The model calls are included, so you need no API key.

What you build
Interview Q&A
samirgenaiacademy.com
Interview Q&A: a trending question with its visual and answer

Scenario questions and mock exams

120 questions · 2 timed mock exams of 60

Twelve exam-style scenario questions for every session, each with the answer and why. Then two full mock exams with a 120-minute timer, flags and a score for each domain.

Unlock the questions and the mocks
Interview coach
samirgenaiacademy.com
Interview Coach: start with your resume

Interview Coach

Resume ▸ job ▸ mock interview ▸ report

Upload your resume and paste any job. It scores your fit, rewrites your resume for that job without inventing anything, then interviews you on it, answered out loud, and gives you a report.

Try a free mock interview
Free questions
samirgenaiacademy.com
Free interview questions and quiz, open to everyone

40 free questions to try first

Free · no payment

Read 40 questions from this course with their answers before you pay, and take the free quiz to check yourself.

Read the free questions
How it works

From sign-up to exam-ready — in one flow

Enroll

Pay once, and access opens immediately. No batch, no timetable.

Build

Sessions 5 to 8: prompts, structured output, tool design and MCP, a lab in each.

›
+
You’re in
Agents understood
Tools and prompts built
Mocks passed.
Exam-ready.

Learn

Sessions 1 to 4: agentic architecture and Claude Code, a narrated video each.

Prove

Sessions 9 and 10: context and reliability, then two timed 60-question mock exams.

1 · Enroll
Pay once, and access opens immediately. No batch, no timetable.
2 · Learn
Sessions 1 to 4: agentic architecture and Claude Code, a narrated video each.
3 · Build
Sessions 5 to 8: prompts, structured output, tool design and MCP, a lab in each.
4 · Prove
Sessions 9 and 10: context and reliability, then two timed 60-question mock exams.
The 10 sessions, in order

The 10 sessions, and what you can do after each

One session a day is a comfortable pace. Each session ends with a lab in three levels: Easy, Moderate and Advanced. Then take the two mock exams.

1Agentic Loops and the Claude Agent SDK. Run the loop on stop_reason, and pick the simplest pattern that workslab
2Multi-Agent Orchestration: Coordinators, Subagents and Hooks. Pass context on purpose, run independent work in parallel, enforce rules with hookslab
3Configuring Claude Code: CLAUDE.md, Rules, Commands and Skills. The right instruction in the right file, loaded at the right timelab
4Claude Code Workflows: Plan Mode, Subagents and CI/CD. Plan the big changes, keep exploration out of the main context, review in CIlab
5Prompt Engineering for Production: Criteria, Examples and Review. Explicit criteria, examples for the unclear cases, and an independent reviewerlab
6Structured Output and Data Extraction: Schemas, Validation and Batches. A schema for the shape, checks for the meaning, batches for the bulklab
7Designing Tools Claude Uses Well: Descriptions, Errors and Choice. Descriptions that pick the right tool, errors an agent can recover fromlab
8MCP Integration: Servers, Resources, Scopes and Secrets. Connect the team to the same servers, keep the secrets out of the repolab
9Context Management: Long Sessions, Case Facts and Large Codebases. Keep the facts that matter, trim what does not, isolate the exploringlab
10Reliability: Escalation, Error Propagation, Provenance and Exam Strategy. Escalate for the right reasons, fail visibly, keep every claim traceablelab
What the scenario questions look like
Agentic Loops and the Claude Agent SDK

Your support agent stops after its first tool call and returns a half answer. The loop checks whether the reply text contains "final answer". What is the right fix?

Keep looping while stop_reason is "tool_use" and finish when it is "end_turn". stop_reason is the signal built for this. Text searches break as soon as the model words things differently.

Multi-Agent Orchestration: Coordinators, Subagents and Hooks

The user told the coordinator to cover only 2024 to 2026. A subagent's findings are from 2019. What is the most likely cause?

The coordinator did not include the date range in the subagent's task; subagents do not inherit the coordinator's conversation. A subagent starts with a fresh context: only its own instructions and the task the coordinator writes for it.

Configuring Claude Code: CLAUDE.md, Rules, Commands and Skills

A senior developer put the team's testing conventions in ~/.claude/CLAUDE.md. A new teammate's Claude Code does not follow them. What is the cause?

The file in the home folder is personal and is never shared through the repository. User-level memory lives only on that developer's machine. Team conventions belong in the project CLAUDE.md or a rules file that is committed.

Claude Code Workflows: Plan Mode, Subagents and CI/CD

A developer must replace the logging library across 45 files, and some modules use it in unusual ways. How should they start in Claude Code?

In plan mode, so Claude explores and proposes a plan that is approved before any file is edited. Large, multi-file work with open questions is where plan mode pays off: a wrong assumption is caught in the plan.

Prompt Engineering for Production: Criteria, Examples and Review

Your automated pull request reviewer is told to "be conservative". Results still vary from file to file. What is the most effective change?

List exactly which kinds of issues to report and which to skip, with what each comment must contain. Vague words give the model no test to apply. Explicit report and skip criteria make the judgement repeatable. Temperature or self-rated confidence do not define what a real issue is.

Structured Output and Data Extraction: Schemas, Validation and Batches

Your pipeline asks Claude to "reply only in JSON". About 1 in 200 replies breaks the parser. Which approach gives reliable structure?

Define a tool with an input schema and have Claude call it, then read the tool input. Tool use with an input schema gives structured input your code reads directly, instead of searching free text. Retries and regex only patch symptoms.

Free · no payment

40 free Claude Certified Architect questions. Try them right now.

Free exam-style practice questions across all five domains, each with the answer and why. Independent practice, not real exam questions. No sign-up, no payment.

Read the free questions ▸ 40 questions · plus a free quiz
The curriculum

10 sessions across the five exam domains

Five domains, two sessions each. Every session has a narrated video, a lab where you write the code, and 12 scenario questions with explanations.

Phase 01Agentic architecture

01

Agentic Loops and the Claude Agent SDK

Run the loop on stop_reason, and pick the simplest pattern that works

Lab: Write an agent loop that is driven by stop_reason, return tool results the right way, and pick the simplest pattern for a task.
02

Multi-Agent Orchestration: Coordinators, Subagents and Hooks

Pass context on purpose, run independent work in parallel, enforce rules with hooks

Lab: Give subagents complete context, run them in parallel, enforce rules with hooks, and check that the whole question is covered.

Phase 02Claude Code

03

Configuring Claude Code: CLAUDE.md, Rules, Commands and Skills

The right instruction in the right file, loaded at the right time

Lab: Work out which instructions apply to a file, and decide which commands are allowed, the way Claude Code does.
04

Claude Code Workflows: Plan Mode, Subagents and CI/CD

Plan the big changes, keep exploration out of the main context, review in CI

Lab: Choose the right way to start a task, and build the CI side of a code review that people actually read.

Phase 03Prompts and output

05

Prompt Engineering for Production: Criteria, Examples and Review

Explicit criteria, examples for the unclear cases, and an independent reviewer

Lab: Turn a vague review prompt into explicit criteria, measure false alarms per category, and build a tagged, two-pass review.
06

Structured Output and Data Extraction: Schemas, Validation and Batches

A schema for the shape, checks for the meaning, batches for the bulk

Lab: Write an invoice schema that allows honest gaps, check the meaning a schema cannot, retry with the exact error, and run batches safely.

Phase 04Tools and MCP

07

Designing Tools Claude Uses Well: Descriptions, Errors and Choice

Descriptions that pick the right tool, errors an agent can recover from

Lab: Write tool descriptions Claude can choose between, return errors an agent can act on, and give each agent only the tools it needs.
08

MCP Integration: Servers, Resources, Scopes and Secrets

Connect the team to the same servers, keep the secrets out of the repo

Lab: Write a team .mcp.json with no secrets in it, pick the transport and scope for each server, and give each subagent only the tools it needs.

Phase 05Context and reliability

09

Context Management: Long Sessions, Case Facts and Large Codebases

Keep the facts that matter, trim what does not, isolate the exploring

Lab: Keep exact facts through a long support conversation, trim tool results to what the task needs, and build a context that puts the facts first.
10

Reliability: Escalation, Error Propagation, Provenance and Exam Strategy

Escalate for the right reasons, fail visibly, keep every claim traceable

Lab: Escalate on clear rules, hand over with a useful summary, pass subagent errors upward with partial results, and keep every claim linked to its source.
Why learners pick it

You prepare by doing, not by reading notes.

Three things make this prep different.

Every domain, in order

Weighted like the reported exam

Agentic architecture, Claude Code, prompts and structured output, tools and MCP, context and reliability: two sessions each.

You write the code

Labs with missing lines

Agent loops, hooks, settings, schemas, MCP config and escalation rules. You write the lines that matter and press Check.

Know when you are ready

Two timed mock exams

60 scenario questions in 120 minutes, flags, and a score for each domain, with every answer explained.

How it compares

Why not just watch free videos?

A fair question. Here is what you get here that you do not get there.

What you get Samir GenAI Academy₹1,999 · 3 months A Udemy course Free YouTube ChatGPT alone
10 sessions across all five exam domains, in order×××
120 scenario questions, every one with the answer and why~×~
Labs where you write agent loops, hooks, schemas and MCP config×~×
10 labs with 114 tasks in your browser, nothing to install~××
Two timed 60-question mock exams with a score for each domain~××
A coach that hears you answer out loud and scores it××~
Try it before you pay: 40 free practice questions and a sample lesson~~
Total cost₹1,999 once · 3 months₹500–3,500 per courseFree, but scatteredMonthly subscription
yes   ~ partly, or you assemble it yourself   × no. Course prices vary with sales; ChatGPT is billed every month, this is one payment.
What early users say

Loved by people who build

One of the few AI learning platforms that focuses on building real projects instead of only theory. The explanations are practical and easy to follow.
Rahul S.Software Developer · via LinkedIn
I was able to understand AI agents and RAG concepts much faster because of the hands-on examples. The content is straightforward and beginner-friendly.
Priya K.Data Analyst · via LinkedIn
The roadmap is well structured and helps avoid information overload. I finally have a clear path for learning Generative AI.
Amit V.Engineering Student · via email
What I liked most is the focus on implementation. Instead of watching endless videos, I could start building immediately.
Neha R.Full-Stack Developer · via LinkedIn
The platform helped me understand how modern AI applications are actually built. The project-based approach makes a big difference.
Karan M.Software Engineer · via email
Feedback early learners sent us on LinkedIn and by email
Claude Certified Architect

Start your exam prep today.

Ten sessions, a lab in every one, then two full mock exams. It starts the moment you pay: no batch, no timetable.

3 months of access

Claude Certified Architect Exam Prep

3-month full access
₹1,999
Inclusive of 18% GST
  • One payment — this does not renew automatically
  • Full access for 3 months from the day you pay
  • Already a student? Use the same email and the course is added to your account
  • No refunds once paid — watch the sample at the top of this page first. Full refund only if we close the course (refund policy)

What’s included

  • 10 sessions, each with a narrated video
  • 10 labs with 114 tasks: you write the code and press Check
  • 120 scenario questions with the answer and why
  • 2 timed mock exams of 60 questions, scored by domain
  • Labs run in your browser, nothing to install
  • Independent prep: our own questions, not real exam questions
  • 3 months of access

Join in 30 seconds

Access opens the moment you pay.

After payment, your login password is emailed to the address above. Please make sure it is correct.
Secure payment via Cashfree
Card details never touch this siteAccess opens immediatelyRefund policy
Credit card · Debit card · UPI · NetBanking
Before you buy

Frequently asked questions

If yours is not here, ask us.

What is the Claude Certified Architect exam prep course?

It is a self-paced online course of 10 sessions that prepares you for the Claude Certified Architect (Foundations) exam: agentic architecture and orchestration, Claude Code configuration and workflows, prompt engineering and structured output, tool design and MCP integration, and context management and reliability. Every session has a narrated video, a lab where you write the code yourself, and 12 scenario questions with explanations, and you finish with two timed 60-question mock exams. It is taught by Samir Kumar, an Enterprise Architect with 21+ years leading AI and engineering teams.

How much does it cost?

The price is ₹1,999, paid once, for three months of access. There is no subscription and no auto-renewal. Until 28 November 2026 there is a launch price of ₹1,499. It is sold on its own and is not part of the three-course bundle.

Are these the real exam questions?

No. The real exam questions are confidential. Every question in the course is our own, written to the domains, weights and scenario style that are publicly reported for the exam, so you practise the same kind of decisions. Samir GenAI Academy is not affiliated with Anthropic.

What does the exam look like?

It is reported to be 60 scenario-based multiple-choice questions in about 120 minutes, with a scaled score where 720 of 1000 passes. Each exam draws on four of six scenarios, such as a customer support agent, a multi-agent research system or Claude Code in CI/CD.

Do I need to install anything?

No. The labs run real Python in your browser. You need a laptop or desktop with a current browser. Reading Python and JSON comfortably is enough to start.

How long do I have access?

Three months from the day you pay. It is one payment and it does not renew by itself.

Can I get a refund?

No. Every purchase is final once you have paid, because the whole course opens on day one. You can watch the sample lesson at the top of this page and try the 40 free practice questions before you decide. You get a full refund if we close the course during your term. The details are in refund policy.

Who made this

The person behind the course

Founded by Samir Kumar, Samir GenAI Academy prepares engineers for the interview that stands between them and an AI role — the concepts, the labs, the questions and the practice.

Samir Kumar

Samir Kumar

Enterprise Architect · Founder and instructor

Enterprise Architect with 21+ years leading AI strategy and engineering teams. He has shipped production GenAI and multi-agent systems on Amazon Bedrock, Azure OpenAI and Google Vertex AI, and built this course so engineers learn by shipping real projects.

  • 16 sessions, including a free AI System Design session
  • 26 narrated videos
  • 19 hands-on labs in your browser
  • 543 interview questions with written answers
Radha Krishna
राधे कृष्ण Every session begins with a prayer. Built with gratitude, taught with patience.
Help & support

Talk to a person

Questions about the course, a payment, or your login? We answer every message, usually within 24 hours on business days.

Email

Write to us

Best for payments and logins: include the email you paid with.

Before you buy

Read the answers

What you get, live or recorded, labs, Python, refunds.

Open the FAQ
Already enrolled?

Log in

Your password is in the email sent after payment. Check spam if it is not there.

Course login
More

Contact page & LinkedIn

Contact us LinkedIn
Ten sessions from now, you will know if you are ready

Pass the Claude Certified Architect exam. In 10 sessions.

Watch, write the code, take the mocks. ₹1,999 · 3 months of access.

Enrol in the exam prep ▸