AI · AIC-107

Forward Deployed Engineer (FDE) 101

Forward Deployed Engineer (FDE) 101 is a hands-on course in customer-facing AI engineering, from finding the real business problem and interviewing stakeholders to planning an AI solution, designing a multi-agent system with Claude Code, connecting APIs and MCP, deploying models on-premise and handing the system over. It suits developers, IT engineers and solution consultants who need to take AI into real use.

Updated 8 Nov 2026

From 5,850 THB / person 6,500 −10% excl. VAT 7% · group rates available

Corporate training quote Join the waiting list

PDFDownload the course outline

Duration18 hours · 3 days
FormatOnsite / live online
Next roundOn request
CertificateIncluded

Course overview

Many organisations have already experimented with AI, yet a lot of projects never get past the demo, because the people who understand the business problem and the people who can build the system sit on opposite sides. A Forward Deployed Engineer (FDE) closes that gap. An FDE works with the customer on site, listens to the real problem, turns a business need into a working AI system and stays until it is genuinely in use, so the role calls for communication, system design and deployment skills in one person.

This intensive course walks learners through the full FDE cycle. Day 1 practises finding the customer's real problem through simulated interviews, planning an AI solution that fits the business and getting started with Claude Code. Day 2 designs the AI architecture: a multi-agent system, API and MCP integration, context isolation with subagents, and turning real requirements into something deployable. Day 3 takes the system on-premise with local AI, sizes the GPU, measures and tunes performance until it meets the SLO, and closes with a handover backed by a runbook and monitoring. (3 days, 6 hours per day, 18 hours in total, Intermediate level.)

What you’ll gain

Understand the role and work cycle of a Forward Deployed Engineer, from discovery to handover
Find the customer's real problem through interviews, process mapping and success metrics
Write an agentic plan and choose an AI solution that fits the problem, the cost and the risk
Use Claude Code to plan, build and test a prototype step by step
Design an AI architecture and a multi-agent system with context isolation through subagents
Connect to customer systems safely through APIs and MCP
Deploy local AI on-premise and calculate GPU, VRAM and KV cache requirements
Measure and tune performance until it meets the SLO, and hand over with a runbook and monitoring

Who this course is for

Developers and IT engineers who want to move into AI work directly with customers or business units
Solution engineers, pre-sales and technical consultants who need to take AI from demo to real use
In-house IT and digital teams that receive AI requests from across the organisation
Tech leads and project leads who take AI systems to production in the cloud or on-premise
People who already use Claude Code or AI agents and want to grow into delivering systems for organisations

Prerequisites

Basic programming in at least one language, such as Python or JavaScript, and the ability to read AI-generated code
Basic Terminal, Git and JSON skills, and a basic understanding of APIs
Your own or your company's Claude plan at Pro level or above, with Claude Code installed before the training day
A laptop that can run Docker Desktop, VS Code and Ollama, with permission to install software

Curriculum

Course Details

This course trains learners in the role of a Forward Deployed Engineer (FDE): an engineer who works with the customer on site, turns a business problem into a working AI system and delivers it until the organisation is using it. It runs for 3 days, 6 hours per day (18 hours in total, 09:00-16:00), as lectures with labs, workshops and role-play built around a single simulated customer case that carries through all three days. Intermediate level. Learners should be able to program at a basic level and use the Terminal and Git, and need their own or their company's Claude plan at Pro level or above, with

Claude Code installed before the training day. The institute does not provide subscriptions, GPUs, servers or cloud credits. Local AI labs run small open models on the learner's laptop, larger models are demonstrated by the instructor, and GPU sizing is taught through calculation exercises. The course builds on Claude Code for Real-World Projects, which focuses on software development with Claude Code, Build AI Multi-Agent with Claude Code, which focuses on directing agent teams, and Mastering AI Agents in the Workplace, which focuses on AI agents for office work. It adds the FDE angle of end-to-end customer delivery, from problem discovery and design to on-premise deployment and handover.

Day 1 Business Discovery and the Agentic Plan

Section 1: Who Is an FDE, and Delivering AI in Organisations

The Forward Deployed Engineer role: working with the customer on site and owning the result until the system is in real use
How an FDE differs from a software engineer, a solution architect and an AI consultant
The FDE cycle: discovery, plan, build, deploy and handover
The skills one person needs: listening, systems thinking, building and talking to executives
Why AI projects get stuck at the demo stage, and how an FDE takes them to production

Section 2: Business Discovery: Finding the Customer's Real Problem

Separate symptoms from root causes with 5 Whys and process mapping
Identify stakeholders: the project sponsor, front-line users, IT and legal
Survey existing data and systems: sources, quality, access rights and PDPA constraints
Set clear business success metrics before writing any code
Lab: draw a process map of a sample workflow and point out where AI can help

Section 3: Workshop: Simulated Customer Interviews (Role-Play)

Prepare a discovery interview guide and open questions that surface real information
Techniques for listening, following up and catching needs the customer does not say out loud
Handle conflicting requirements, unrealistic expectations and budget or data constraints
Role-play: take turns as the FDE and the customer on a simulated business case
Turn the interview into a problem statement and review it together

Section 4: The Agentic Plan: An AI Solution That Fits the Business

Choose the right level of solution: prompt, workflow, RAG, single agent or multi-agent
Build or buy, cloud API or local model, weighed against cost, speed and data confidentiality
Assess the risks: hallucination, data security and where a human must review
Write a solution brief and set the scope of a measurable pilot
Workshop: write a one-page agentic plan from the interview results

Section 5: Lab: Getting Started with Claude Code

Start Claude Code in the Terminal or the Code tab of Claude Desktop and set its permissions
Plan Mode: let Claude analyse the project and propose a plan before acting
CLAUDE.md for project context, team conventions and the customer's business vocabulary
Work step by step, review diffs and use Git as a rollback point
Lab: have Claude Code scaffold a prototype from the agentic plan

Day 2 AI Architecture for Real Projects

Section 6: Designing an AI Architecture from Requirements

Turn the problem statement into functional and non-functional requirements
The building blocks of an AI system: model, context, tools, memory, data and guardrails
Draw architecture diagrams and data flows that both the customer and IT can follow
Record the reasoning behind each decision in an Architecture Decision Record (ADR)
Lab: design the architecture for the customer case from Day 1

Section 7: Designing a Multi-Agent System

When a multi-agent system is worth it, and when one agent is enough
Common patterns: orchestrator-worker, pipeline, router and reviewer
Split agent responsibilities by scope and define clear inputs and outputs
The extra token cost and latency that come with more agents
Lab: sketch the agent team for the customer case with each agent's role

Section 8: Lab: Context Isolation with Subagents

Context overflow and mixed-up information in long single sessions
Subagents with their own context that return only a summary to the main agent
Give each subagent only the prompt, tools and permissions it needs
Keep confidential customer data away from agents that do not need to see it
Lab: build subagents for research, coding and review in Claude Code

Section 9: Lab: Connecting APIs and MCP

Call the customer's REST APIs safely: authentication, rate limits and error handling
The Model Context Protocol (MCP): servers, tools and resources
Connect existing MCP servers to Claude Code and restrict what they can do
Build a small MCP server that wraps an internal API for agents to use
Store secrets properly and never embed API keys in code or prompts

Section 10: Workshop: From Requirements to a Deployable System

Break the architecture into tasks and let Claude Code build them one at a time
Write tests and an evaluation set (eval set) from the customer's sample cases
Package the system with Docker and per-environment configuration files
Demo to the simulated customer, gather feedback and adjust the plan before going live

Day 3 Local AI Deployment and Production Readiness

Section 11: Lab: Deploying Local AI On-Premise

Why organisations choose on-premise: data confidentiality, regulation, cost and closed networks
Choose an open model that fits the task and check its licence before commercial use
Ollama for single machines and prototypes, vLLM for servers with GPUs
Expose an OpenAI-compatible API so applications and agents can call the internal model
Lab: run a small model on your laptop with Ollama and connect it to the Day 2 system

Section 12: Workshop: GPU and Hardware Sizing

VRAM needed for a given model size and numeric precision (FP16, INT8, 4-bit)
Quantisation: what you gain, what you lose and how to pick the right level
KV cache: calculating it from context length and the number of concurrent users
Plan for concurrency and choose between one GPU, several GPUs or several machines
Workshop: size the hardware for the customer case and write a hardware proposal

Section 13: Lab: Measuring and Optimising Until the SLO Is Met

Agree SLOs with the customer: latency, time to first token, throughput and error rate
Measure tokens per second and latency under load with test scripts
Tune model size, quantisation, batching, context length and prefix caching
Measure answer quality alongside speed with the same eval set
Lab: load-test the system, read the results and tune until it meets the SLO

Section 14: Production Readiness and Handover

A production checklist: security, access control, logging, backup and a rollback plan
Monitoring: dashboards for latency, GPU usage, errors and answer quality
Write a runbook for common incidents and the escalation steps
Handover documents, user training and acceptance criteria
Ongoing care: model updates, re-evaluation and measuring business results after go-live

Section 15: Capstone: Delivering an FDE Project End to End

Combine three days of work into a delivery pack: problem statement, agentic plan, architecture and a running system
Present to simulated customer executives on business value, cost and risk
Answer technical questions and handle objections professionally
Review together, get feedback from the instructor and plan your own path into the FDE role

Schedule & training options

For individuals — public rounds

No public rounds are open right now. Join the waiting list and we will contact you first when the next round opens, or ask us on LINE. Or call 02-570-8449 or 088-807-9770

Join the waiting list Ask about rounds

For organisations — in-house / private

Tailor the content to your team’s tools and projects
Your dates, at your office or live online
Quotation with tax ID for procurement

Corporate training quote

Instructors

Samit KoyomFounder & CEO, IT Genius Engineering

Frequently asked questions

How does this course differ from Claude Code for Real-World Projects, Build AI Multi-Agent with Claude Code and Mastering AI Agents in the Workplace?

This course builds on all three. Claude Code for Real-World Projects focuses on software development with Claude Code from issue to pull request, Build AI Multi-Agent with Claude Code focuses on building and directing agent teams, and Mastering AI Agents in the Workplace focuses on no-code AI agents for office work. This course takes the FDE angle of end-to-end customer delivery: problem discovery and customer interviews, the agentic plan, architecture design, then on-premise deployment, GPU sizing, measuring against SLOs and handover.

Which accounts and software do I need, and does the institute provide them?

The institute does not provide accounts or subscriptions. You use your own or your company's Claude plan at Pro level or above, such as Pro, Max, or a Team or Enterprise seat that includes Claude Code, and install Claude Code before the training day. Pro works for the course but may reach its usage limit during long labs. You should also install Git, VS Code, Docker Desktop and Ollama on your laptop in advance.

Do I need a machine with a GPU, and does the institute provide GPUs or servers?

No GPU is required, and the institute does not provide GPUs, servers or cloud credits. Local AI labs run small open models on your own laptop with Ollama, or vLLM where your hardware allows. Larger models are demonstrated by the instructor on more powerful hardware, and GPU, VRAM and KV cache sizing is taught through calculation exercises you can apply to real hardware plans.

I am not a full-time developer. Can I take this course?

Yes, as long as you can program at a basic level in at least one language, use the Terminal and Git, and read code that AI generates. Most labs let Claude Code write the code, but you still review it and make the decisions. The course suits IT engineers, solution engineers and technical consultants with that background. If you have never written code, start with a no-code AI agent course first.

What certificate do I receive at the end?

Learners who attend the full course receive a Certificate of Completion from IT Genius Institute. It is not a certification from any tool vendor. The course focuses on practical skills and on the capstone delivery pack, which you can use as a portfolio example.

Can I bring a real company problem to class, or run the course in-house?

In public classes everyone works on the same simulated customer case. You can adapt ideas from your own problem during the workshops, but do not bring confidential or personal company data into class. If you want to work on your organisation's own problem and systems, the course can be run in-house and you can request a quotation.