Developers shipping their first agent
You can already build software. You want the patterns that keep an agent stable once real users touch it.
LevelIntermediate
AI Engineering with Claude
For developers who want to ship AI agents — not just demo them
- Build on the same Claude tools real teams ship with
- Four portfolio projects taken from real industry problems
- Watch at your own pace — stuck on something? Message me any time
- 38
- hours of video
- 4
- modules
- 39
- lessons
- 4
- projects
I use this stack every day
This is not theory off a slide
I have shipped software for 13 years in Hong Kong, Germany and Japan, and I build with Claude Code and the Agent SDK on real work.
So this course teaches what actually survives production — the loop, the tools, the evaluation and the guardrails that hold.
Who it's for
Built for developers moving into AI engineering
Backend and platform engineers
You are being asked to put AI into an existing system, and you want it observable, evaluated and governed.
Tech leads choosing an approach
You need to judge model choice, tool design and autonomy limits before you commit a whole team to them.
Before you start
This is an intermediate course
It assumes you can already write and run code. If several of these are new to you, start with the Full-Stack Bootcamp instead.
- Comfortable with basic Python
- Comfortable with basic TypeScript
- Have called an LLM API at least once
- Know roughly what an AI agent is
Skills
Ten things you will be able to do
Each one is taught against a system that runs, not a slide that explains.
- Claude Code platform
- Claude API
- Agentic workflows
- Claude Agent SDK
- MCP architecture and roles
- MCP-integrated chatbots and agents
- AI agent observability
- AI agent evaluation
- Agentic AI governance
- Multi-agent orchestration
Curriculum
Four modules, 39 lessons, four projects
Every module ends in a milestone project you run, break on purpose, and then defend.
Tap a module to see every lesson
Put Claude to work on real systems: choose the right model for the job, design the agent loop, and configure Claude Code for a whole team.
Tools used
Claude APIClaude Agent SDK
Claude CodeClaude Skills
Claude APIClaude Agent SDKClaude CodeClaude Skills- 1Set up your workspace and API keys
- 2The Claude 4.5 model family
- 3Choosing a model with the Claude API
- 4Introduction to agentic system design
- 5Designing agent architectures
- 6Introduction to the Claude Agent SDK
- 7Building production agents with the Agent SDK
- 8Build a claims intake agent on a stop_reason loop
- 9Engineer a context strategy for a long-running support copilot
- 10Introduction to Claude Code
- 11Configuring Claude Code
- 12Introduction to Claude Skills
- 13Creating your own Claude Skills
- 14Configure Claude Code for a multi-surface monorepo team
- 15Build a multi-shift quality monitoring system
Project
Four working systems, defended
Stand up, run and verify four reference systems from this module, then defend every design decision with your own run output.
Connect AI applications to the tools, data and systems they need through the Model Context Protocol — then govern exactly what they are allowed to touch.
Tools used
Model Context ProtocolFastMCPClaude Agent SDK
Python
Model Context ProtocolFastMCPClaude Agent SDKPython- 1Introduction to MCP in Action
- 2MCP fundamentals
- 3MCP architecture and roles
- 4Server features: tools, resources and prompts
- 5Client features: roots, sampling and elicitation
- 6Building MCP servers with FastMCP
- 7Integrating MCP servers with Agent SDK agents
- 8Build a retail inventory agent with MCP tools
- 9Introduction to custom tools in the Agent SDK
- 10Implementing custom tools in the Agent SDK
- 11MCP governance: scoped config and an audited agent loop
Project
PriceScout, an agentic analyst
Write a client that commands your own scraper and database servers to gather competitor pricing and reason over it.
Measure what an agent actually does and watch how it behaves in production, so silent failures surface before your users find them.
Tools used
Zod
TypeScriptClaude APIPython
ZodTypeScriptClaude APIPython- 1Introduction to evaluation and observability
- 2Introduction to structured outputs
- 3Implementing structured outputs with schemas
- 4Build a resilient mortgage document extraction system
- 5Agent evaluation strategies
- 6Evaluate Agent SDK agents with traces and test cases
- 7Build a validated, routed insurance policy pipeline
- 8Investigate supply chain risk with multi-source synthesis
Project
Break your own systems
Operate the three document and risk systems you built, break them on purpose, and defend their design with evidence from your own runs.
Give agents room to act on their own while keeping them inside firm boundaries — ones enforced by code, not by a well-worded prompt.
Tools usedClaude Agent SDKClaude CodePython
Claude Agent SDKClaude CodePython- 1Introduction to bounded autonomy and guardrails
- 2Introduction to multi-agent orchestration
- 3Orchestrating multi-agent systems with the Agent SDK
- 4Build a hub-and-spoke multi-agent system
- 5Enforce agent compliance with deterministic hooks
Project
Enterprise code-review orchestrator
Build a code-review tool that routes work across several agents, each with its own tools, and holds every one of them accountable.
Projects
Four projects for your portfolio
Each is drawn from a real industry problem — insurance, retail, financial services and manufacturing.
Module 1
Four working systems, defended
Stand up, run and verify four reference systems from this module, then defend every design decision with your own run output.
Prove you can run and defend an agent architecture
Module 2
PriceScout, an agentic analyst
Write a client that commands your own scraper and database servers to gather competitor pricing and reason over it.
Prove you can build and command your own MCP servers
Module 3
Break your own systems
Operate the three document and risk systems you built, break them on purpose, and defend their design with evidence from your own runs.
Prove you can measure an agent instead of trusting it
Module 4
Enterprise code-review orchestrator
Build a code-review tool that routes work across several agents, each with its own tools, and holds every one of them accountable.
Prove you can keep several agents accountable at once
Stack
The tools you will actually use
No framework tourism. One coherent stack, carried across all four modules.
Claude platform
Claude 4.5
Pick Haiku, Sonnet or Opus by intelligence, speed and cost
Claude API
Direct model calls, tool use and structured outputs
Claude Code
An agentic assistant that reads and writes your codebase
Claude Agent SDK
Production agents with goals, tools and permissions
Claude Skills
Reusable expertise your agents can discover and apply
Protocol and tooling
Model Context Protocol
The standard way agents reach tools, data and systems
FastMCP
Build your own MCP servers quickly
Custom Tools
Wire agents into private data and internal APIs
Languages and validation
Python
Servers, governance layers and evaluation harnesses
TypeScript
Agent SDK work and typed tool definitions
Zod
Schemas that stop an agent inventing values
How it works
Self-paced, with a person on the other end
Enrol on LINE
Message me first. If the course is wrong for you, I will say so.
Get access within 1–2 days
Your platform login and Discord invite are issued once payment clears.
Watch at your own pace
38 hours across four modules, all unlocked from day one and yours for a full year.
Build the projects
Four milestone projects, every one of them run against the real Claude API.
Send your code for review
I read it myself and reply with what I would change and why.
The platform
Where you will actually watch
The same learning platform as our bootcamp: progress tracking, notes, and playback that resumes where you left off on any device.
Access guaranteeOne full year of access, counted from the day your account is issued.
Outcomes
What you can ship afterwards
Not a certificate on its own. Working systems, and the judgement to defend them.
- An agent loop that decides on its own when to clarify, route or escalate
- MCP servers your team can point any client at
- Structured outputs that refuse to fabricate values
- An evaluation harness that catches regressions before users do
- Deterministic hooks that enforce rules a prompt cannot
- A multi-agent system with a clear handoff contract
Every one of these comes out of a project you built yourself.
What Varis Lab students say
ViVian
Finally Finished a Course I Could Keep Up With
"I had studied elsewhere but couldn't keep up and never finished. Studying with Instructor Varis in a group of 4, whenever I had questions or fell behind, I could ask immediately — it helped so much. This time I actually completed the course."
• Small group of 4 — questions answered on the spot
• Finished the whole course without falling behind
Nadthaporn
Hands-on Learning with Real Projects
"It's not just theory — this course lets you work on real projects. I felt like I gained actual work experience. I practiced every step of the process."
• Completed the course and built a portfolio from the projects
• Confident — I have the knowledge and skills the market demands
Pensri
Coding Every Day Until It Finally Clicked
"Coding with the instructor every day until course completion — now I can code fluently on my own instead of copying from tutorials."
• Focused on Front End, but comfortable with Back End tasks too
• Daily practice turned into real coding fluency
Stang
Fun Teaching Style with Clear Examples
"This course made complex topics feel simple. The instructor is great at Back End, explains well with examples, making it easy to understand the concepts and start coding right away."
• Confident in my own skills
• Ready to apply for developer jobs abroad
Dancio
Stopped Wasting Time Guessing What to Study
"I wasted so much time studying on my own before finding Instructor Varis's course. Honestly, I wish I had decided to join much sooner."
• A clear learning path instead of guessing what comes next
• Covered both Front End and Back End, not just one side
Me
Learned the Skills Real Projects Actually Need
"In this course, we built APIs and managed databases with real projects — exactly the kind of work I wanted to be able to do on my own."
• Hands-on practice in every class
• Started with zero experience and still kept up
Your Instructor
Varis Darasirikul
Founder & CEO of Varis Lab
Over the past decade, I've built my career as a Software Engineer, working across 4 countries (Hong Kong, Germany, Japan, Thailand) with leading corporations and startups. I hold a Master's degree in Computer Science from the University of Hong Kong, and today I'm the CEO & Founder of Variden Tech, building AI-powered products and software. 🇹🇭 🇭🇰 🇩🇪 🇯🇵
I've been on both sides of the table — as a candidate facing tough interviews and as part of teams hiring globally. This gives me unique insights into what international recruiters actually look for.
10+
Years Experience
4
Countries
Master's in CS
Univ. of Hong Kong
The University of Hong Kong
Tuition
Ship production AI agents, for only
One-time payment
25,000
THB / person
Or pay over 10 months
2,500
THB / month
0% interest for up to 10 months with credit cards from participating banks.
Instalments available with these cards
What the price includes
- 38 hours of video across four modules, with 1 year of access
- 39 lessons and 4 milestone projects
- Code Review by a Senior Developer
- Instructor support throughout the course via Discord
- Discord community membership
- Starter repositories and prompts for every project
- Free access to future updates of this course
FAQ
Questions before you enrol
No. You need to be able to write code. The course starts from how the Claude models differ and builds up from there.
Both. Python for MCP servers and governance layers, TypeScript for Agent SDK work and typed tools. Basic familiarity with each is enough.
One full year from the day your account is issued, and there is no limit on how many times you rewatch a lesson.
No. The bootcamp teaches you to build software. This course assumes you already can, and teaches you to build agents that survive production.
Yes, a small amount. The projects call the real Claude API, which is the point. Typical usage across the whole course is modest, and I show you how to keep it that way.
Taught in Thai, with English kept for technical terms, code and documentation.
Yes, on Discord, for your whole access period. I answer them myself.
Yes. Send a repository, I read it, and I reply with what I would change and why.
I update the affected lessons, and the updates are free for anyone already enrolled.
Yes, once you finish the course and submit the projects. In interviews, though, what carries weight is the four working systems you can show and explain.
Ready to build agents that hold up?
Message me on LINE. Tell me what you are building and I will tell you honestly whether this course fits.
Cancellation Policy
Cancellation & Refund Policy
Terms & Conditions