Forward Deployed Engineer: Foundations and Advanced
10 short courses in 3 levels, 43 core lessons and about 17 hours. Build the service in Python, ship it into a customer's environment, prove it works, then practise the interview. Each short course is small enough to finish in a few sittings: take them in order or pick the one you need. Pairs with the AI Architect course and the AI Solution Architect course.
The 3 levels
Open one to see what you finish with and every lesson in it.
- Foundations about 8.8 h26 lessons
FDE Foundations: Build an AI Service
Python and FastAPI, LLM features and retrieval: a tested service you can explain.
3 short courses: Python and FastAPI · LLM features · RAG and data
Not started · 5 optional
- Advanced about 6.7 h21 lessons
FDE Advanced: Ship It and Prove It
Agents and guardrails, integrating and shipping into a customer's estate, and an evaluation report you can defend.
3 short courses: Agents and guardrails · Integrate and ship · Evaluate and capstone
Not started · 4 optional
- Interview and career about 3.4 h17 lessons
FDE Interview and Career: Land the Role
Scored design exercises, question banks for technical and customer rounds, and a portfolio and outreach plan.
4 short courses: Framework and exercises · Technical questions · Customer questions · Career positioning
Not started · 3 optional
FDE Foundations: Build an AI Service
26 lessons (21 core, 5 optional) · about 8.8 h
Part 1Forward Deployed Engineer: Python and FastAPI Foundations
The role, your baseline, and a typed, tested FastAPI service in Python for Java and Node engineers.
9 lessons · about 3.6 h
You finish with
- A written 30/60/90 plan and a gap map from your own experience.
- A tested AI-ready API skeleton: product catalog, typed models, search endpoint.
Lessons
- 1.Forward Deployed Engineer Role Compared With Related Roles10 minStart here
- 2.Enterprise Integration Skills for FDE Roles: A Baseline10 min
- 3.FDE Capstone and 30/60/90 Plan for Enterprise AI10 min
- 4.Python for Java and Node Engineers: Syntax, Types, venv20 min
- 5.Dataclasses and Pydantic: Validate Data at the Boundary20 min
- 6.FastAPI Service for AI Apps, Compared with Spring Boot25 min
- 7.Async, HTTP Clients, Timeouts and Retries in Python20 min
- 8.Errors, Configuration and Logging in a FastAPI Service15 min
- 9.Testing with pytest: Fixtures, Parametrize, FastAPI20 min
Part 2Forward Deployed Engineer: Build LLM Features
An LLM feature built like an engineer would: untrusted input, structured output, a provider boundary, streaming and one tool.
7 lessons · about 2.5 h
You finish with
- A structured-response endpoint with a mock provider and one read-only tool.
Lessons
- 1.How an LLM Application Works: One Request in Code15 minStart here
- 2.Prompts and Untrusted Input: A First Injection Test20 min
- 3.Structured Outputs: JSON Schema and Pydantic Validation25 min
- 4.LLM Provider Calls: Adapter, Mock, Rate Limits, Cost30 min
- 5.Tool Calling: One Read-Only Tool and Argument Validation20 min
- 6.Streaming Responses and Invalid Model Output10 min
- 7.Real Model Lab: Variability, Failures and LatencyOptional21 min
Part 3Forward Deployed Engineer: RAG, Search and Data
Document ingestion, chunking, keyword and vector search, access control, citations, evaluation, and keeping the data current.
10 lessons · about 2.7 h
You finish with
- A RAG service with cited answers and measured retrieval results.
Lessons
- 1.Document Ingestion and Metadata for RAG20 minStart here
- 2.Chunking Trade-offs Measured on a Fixed Question Set20 min
- 3.Keyword, Vector and Hybrid Search for RAG25 min
- 4.Metadata Filters and Tenant-Aware Access Control in RAG20 min
- 5.Context Assembly and Citations for RAG Answers20 min
- 6.RAG Evaluation: Recall, Groundedness and Failure Modes15 min
- 7.Embeddings and pgvector Tuning for RAGOptional30 min
- 8.Reranking: A Second Stage for RAG RetrievalOptional20 min
- 9.Keep RAG Data Current: Sync, Deletes, Messy FilesOptional22 min
- 10.Prompt, RAG or Fine-Tune: A Decision GuideOptional18 min
FDE Advanced: Ship It and Prove It
21 lessons (14 core, 4 optional) · about 6.7 h
Part 1Forward Deployed Engineer: Agents, Tools and Guardrails
Bounded workflows, validated tool calls, permissions, human approval, failure tests, an MCP server, guardrails and caching.
8 lessons · about 2.7 h
You finish with
- A bounded workflow with two read-only tools, step limits and an approval gate.
Lessons
- 1.Workflow or Agent: Choosing Control Flow in Code15 minStart here
- 2.Tool Definitions and Validated Arguments in Python25 min
- 3.Tool Permissions: The Application Enforces, Not the Prompt25 min
- 4.Workflow Limits, Step Log and Resumable State in Python20 min
- 5.Human Approval Gate for AI Agent Actions in Python15 min
- 6.Failure Injection Tests for a Tool-Calling Workflow20 min
- 7.An MCP Server for the Product CatalogOptional30 min
- 8.LLM Output Guardrails and Tenant-Safe Response CachingOptional20 min
Part 2Forward Deployed Engineer: Integrate and Ship
Where the service sits in the customer's estate, passing identity, reliable calls, the design note, and shipping into their environment.
5 lessons · about 1.4 h
You finish with
- An integration design note for the capstone.
Lessons
- 1.Where the Python AI Service Sits in the Estate15 minStart here
- 2.Passing Identity: Authn, Authz and Tenant Isolation20 min
- 3.Timeouts, Retries, Idempotency; Sync or Async15 min
- 4.The Integration Design Note, and When Java Wins10 min
- 5.Ship the AI Service into a Customer's EnvironmentOptional32 min
Part 3Forward Deployed Engineer: Evaluate and Capstone
A golden set, a baseline against strategies, cost and failures, the evaluation report, and the capstone brief, contract and rubric.
8 lessons · about 2.6 h
You finish with
- An evaluation report you could hand to a customer.
Lessons
- 1.Golden Set for AI Evaluation: Build One by Hand15 minStart here
- 2.Baseline vs Strategy Evaluation with a Golden Set20 min
- 3.AI Latency, Token Cost and Failure Capture per Run15 min
- 4.AI Evaluation Report Template: Method, Results, Limits10 min
- 5.Capstone Brief: Requirements, Dataset and PhasesCapstone8 min
- 6.Capstone Architecture and API ContractCapstone8 min
- 7.Capstone Rubric (100 Points) and SubmissionCapstone8 min
- 8.Model-Graded Evaluation: Judge the Judge FirstOptional20 min
FDE Interview and Career: Land the Role
17 lessons (8 core, 3 optional) · about 3.4 h
Part 1FDE Interview: Answer Framework and Design Exercises
An answer framework, three scored design exercises, a timed round and a capstone demo script.
6 lessons · about 1 h
You finish with
- Scored answers to three design exercises and six timed questions.
Lessons
- 1.FDE Interview Loop and Answer Framework10 minStart here
- 2.Enterprise RAG Design Exercise With a Scored Answer12 min
- 3.Agentic Workflow Design Exercise With a Scored Answer12 min
- 4.AI Gateway and Integration Design Exercise12 min
- 5.Timed Interview Round: Six Core Questions14 min
- 6.Capstone Demo Script and Interview PresentationCapstone8 min
Part 2FDE Interview: Technical Question Banks
Python and API, RAG and agents, and security and reliability questions, each with a strong answer and a rubric.
5 lessons · about 1.2 h
Lessons
- 1.Python and API Interview Questions: FastAPI and AsyncQuestion bank10 minStart here
- 2.RAG and Agent Interview Questions: Retrieval to ApprovalQuestion bank10 min
- 3.Security and Reliability Interview Questions for AI AppsQuestion bank10 min
- 4.Extended Interview Bank 2: Python, RAG and AgentsOptional20 min
- 5.Extended Interview Bank 3: Security, Reliability, BehavioralOptional20 min
Part 3FDE Interview: Customer and Behavioral Questions
Customer scenarios and behavioral questions, each with a strong answer and a rubric.
3 lessons · about 0.7 h
Lessons
- 1.FDE Scenario Interview Questions: Four Customer CasesQuestion bank10 minStart here
- 2.Behavioral Interview Questions: Ownership, Failure, ConflictQuestion bank10 min
- 3.Extended Interview Bank 1: FDE and Solution ScenariosOptional20 min
Part 4FDE Career: Positioning, Portfolio and Outreach
Present your enterprise experience, build a portfolio and a short profile, and keep a weekly practice routine.
3 lessons · about 0.5 h
You finish with
- Two CV variants, a portfolio case study and a weekly tracker.
Lessons
FDE Foundations: Build an AI Service
Python and FastAPI, LLM features and retrieval: a tested service you can explain. About 8.8 h: 3 short courses, 21 core lessons, 5 optional.
- Build a typed, tested Python API that wraps an LLM with structured output and a mock provider.
- Treat every prompt as untrusted input and write a first injection test.
- Build and measure retrieval: chunking, keyword and hybrid search, access control inside the query, and citations.
- Keep retrieval data current with real deletes, and decide what to do with scans and tables.
| Module | You finish with | Lessons | Hours | Capstone |
|---|---|---|---|---|
| M0: Career orientation and baselineKnow what the role is, where you start, and how the course and capstone work. | A written 30/60/90 plan and a gap map from your own experience. | 3 | 0.5 h | - |
| M1: Python and FastAPI for enterprise engineersPython, typed models, FastAPI, async HTTP, errors and pytest, explained for Java and Node engineers. | A tested AI-ready API skeleton: product catalog, typed models, search endpoint. | 6 | 2 h | 1.1 h |
| M2: LLM application engineeringBuild an LLM feature the way an engineer would: untrusted input, structured output, a provider boundary and one tool. | A structured-response endpoint with a mock provider and one read-only tool. | 6 | 2 h | 0.5 h |
| M3: RAG: build and evaluateBuild and measure retrieval: ingestion, chunking, search, access control, context and citations. | A RAG service with cited answers and measured retrieval results. | 6 | 2 h | 0.7 h |
FDE Advanced: Ship It and Prove It
Agents and guardrails, integrating and shipping into a customer's estate, and an evaluation report you can defend. About 6.7 h: 3 short courses, 14 core lessons, 4 optional.
- Build a bounded agent workflow with validated tools, step limits, a step log and human approval.
- Place a Python AI service in an enterprise estate, pass identity and tenancy across the boundary, and ship it into a customer's environment.
- Write an evaluation report with a golden set, a baseline, two strategies, cost and failures.
| Module | You finish with | Lessons | Hours | Capstone |
|---|---|---|---|---|
| M4: Bounded agent workflowsA bounded workflow with validated tools, permissions, limits, a step log and a human approval gate. | A bounded workflow with two read-only tools, step limits and an approval gate. | 6 | 2 h | 0.7 h |
| M5: Integrating a Python AI service into an enterprise estateWhere a Python AI service sits in an enterprise estate, how identity and tenancy cross the boundary and when Java wins. | An integration design note for the capstone. | 4 | 1 h | 0.4 h |
| M6: Hands-on evaluation reportMeasure your system honestly: a golden set, a baseline, two strategies, cost, latency and failures. | An evaluation report you could hand to a customer. | 4 | 1 h | 1.1 h |
The capstone is built in short phases inside the modules; its build time is listed in the Capstone column and is part of the hours, with 0.5 h kept for review. If you already ship LLM features you can skip modules one and two.
FDE Interview and Career: Land the Role
Scored design exercises, question banks for technical and customer rounds, and a portfolio and outreach plan. About 1.5 h: 4 short courses, 8 core lessons, 3 optional.
- Answer design and scenario interview questions with a framework, and score your own answers.
- Prepare for technical and customer rounds with question banks that say what is assessed.
- Present your enterprise experience, your capstone and a short profile, and keep a weekly practice routine.
| Module | You finish with | Lessons | Hours | Capstone |
|---|---|---|---|---|
| M8: Interview preparationInterview preparation: the loop, an answer framework, three design exercises and a timed round. | Scored answers to three design exercises and six timed questions. | 5 | 1 h | - |
| M9: Career positioningPosition your enterprise experience, build a portfolio entry and keep a practice routine. | Two CV variants, a portfolio case study and a weekly tracker. | 3 | 0.5 h | - |
How the lessons work
- Every lesson opens with one overview figure, then explains with diagrams, tables and short text. Architecture figures are animated.
- Code is never retyped: every Python sample is an excerpt of a file that a test runs, or is labelled Pseudocode.
- Each lesson ends with an exercise with acceptance criteria and staged hints, and a five-question check that explains why.
- Each lesson has an In the interview or At the customer note, so the skills carry into the role and the loop.
Start with what a Forward Deployed Engineer does .
Which course should I take?
They are written to be used together: the Architect course teaches the design judgement, the Forward Deployed Engineer course teaches you to build and deliver, and the AI Solution Architect course teaches you to scope and propose it with a customer. Lessons link across and never repeat each other.
| Topic | AI Architect | Forward Deployed Engineer (this course) | AI Solution Architect |
|---|---|---|---|
| The question it answers | How should this AI system be designed, reviewed, operated and governed? | How do I build it in Python, ship it into a customer's environment and prove it works? | How do I turn a customer's business problem into an approved design, estimate and proposal? |
| What you produce | Requirements, decision records, diagrams, design reviews, governance controls. | A tested Python service, an evaluation report, an integration design note, and a shipped, checked deployment. | A design pack: outcomes, scope, security answers, an estimate with total cost, a proposal and a review defense. |
| How much code | Little. Mechanisms are explained, not built. | A lot. Every code lesson runs in a test suite. | None. It is about decisions, documents and conversations. |
| Format | Two courses and 41 case files. | Three levels in 10 short courses, about 17 hours, a capstone and interview question banks. | Two levels in 3 short courses, about 6 hours, one running project. |
| Start here if you are | A tech lead, architect or manager who signs off on AI designs. | An integration or backend engineer moving into customer-facing AI work. | A consultant, pre-sales engineer or engineer who has to scope and propose AI work. |
Not sure? If you already design systems and want to build them, start with the Forward Deployed Engineer course. If you already build and want the design judgement, start with the AI Architect course. If your job is to scope and propose AI work, start with the AI Solution Architect course.
In the course
- Capstone briefThe Enterprise Product Discovery Copilot: requirements, dataset and the eight phases.
- Capstone rubricThe 100-point rubric and what to submit.
- Interview question bankQuestions with what is assessed, an answer framework, a strong example and a scoring rubric.
- What a Forward Deployed Engineer doesThe role compared with applied AI engineer and AI architect.
- Build: practice quizShort checks across the build modules.
- Build: practice examScenario questions with revision links for every miss.
- Interview and career: practice quizShort checks on the answer framework and positioning.
Questions
- Who is this course for?
- Backend and integration engineers, for example Java and Spring Boot or Node.js developers with two or more years of experience, who are new to Python and to building AI features and want to work in customer-facing AI engineering roles.
- Where should I start?
- Start with the first Foundations course, Python and FastAPI. If you already ship LLM features you can skip the first two modules. Each short course takes a few sittings, and the first lesson of each is open to everyone. Take Advanced once the Foundations service works.
- What is the difference between Foundations and Advanced?
- Foundations builds the service: Python and FastAPI, LLM features and retrieval. Advanced covers what a customer deployment needs: agents and guardrails, integration and shipping, and the evaluation report and capstone. Interview and career is the last level: design exercises, question banks and your positioning.
- How is it different from the AI Architect course?
- The AI Architect course teaches how to design, review and govern an AI system. This course teaches how to build one in Python and deliver it to a customer. The lessons link across and do not repeat each other.
- How long does it take?
- About 17 h in total: Foundations about 8.8 h, Advanced about 6.7 h and Interview and career about 1.5 h. The optional lessons and the interview question banks are not counted. With the AI Solution Architect course the whole path is about 23 hours.
- Do I need a paid API key?
- No. The capstone runs against a deterministic mock model provider, and every test in the course runs without a model key. One real provider adapter is included, with a lab that shows what changes when the model is real, if you want to try it with your own key.
- Is it free?
- Yes. The first lesson of each short course is open to everyone, and a free account unlocks the rest, tracks your progress and keeps your place.
- Will it get me a job?
- No course can promise that. It teaches the engineering and the interview reasoning the role asks for, and gives you a working project and a written evaluation to show. It does not claim that completing it makes a system production-ready, and it contains no salary or hiring data.
- Are the customer scenarios real?
- No. Every customer scenario is labelled "Composite scenario, not a real company" and its numbers are illustrative. The capstone's catalog and documents are fictional.
- How do I know the facts are right?
- Claims about libraries and services are checked against the official documentation with the date they were checked, the arithmetic is recomputed, and every verified code sample is an excerpt of code that runs in the capstone's tests. The pages were drafted with AI assistance and reviewed by a second pass; none carries the author's Reviewed badge yet.