Learning Path: From Zero to Your First Project ​

In this chapter you will learn: how to choose a learning order based on your background, what to read, do, and deliver each week, and when to stop and shore up your fundamentals.

First, Make a Choice ​

"Zero foundation" can mean very different starting points. Spend 10 minutes on the self-assessment, then find your match:

Below, 12 weeks is the default pace. 6-8 hours per week, ideally split as:

2 小时  读章节 / 看资料
3 小时  动手做（Lab、练习、访谈）
1 小时  写复盘 / 更新作品集
1 小时  缓冲（补没看懂的地方）

If you can only invest 3 hours per week, stretch the 12 weeks to 24 — don't compress the exercises.

12-Week Overview ​

This is a "minimal closed loop" path, not the only one. You can start the Lab in week 6 and go back to Chapter 4 later; or read Chapter 9 first and work backward from interview questions to identify skills to fill.

Phase One: Awareness and Foundations (Weeks 1-4) ​

Goal: Be able to explain FDE in your own words, and independently complete one API call, one SQL query, and one Git commit.

Week 1: Build concepts

• Read Chapter 1, focusing on "how the FDE differs from eight adjacent roles."
• Write 300 words: If you had to explain FDE to someone with zero technical background, what would you say?
• Common mistake: memorizing the definition without understanding "why this role is needed." Fix: read the three gaps in Chapter 2.

Week 2: Understand the field

• Read Chapter 3 and record what an FDE actually does in a typical week.
• Find 3 "AI deployment" scenarios around you (your company, news, friends' projects) and apply the five field questions: Who hurts? How badly? How is it solved today? Why now? Who pays?
• Deliverable: a three-row table, one scenario per row, marked "worth doing / not worth doing / insufficient information."

Week 3: Tech foundations I

• Read the first half of Chapter 4: computers, networking, APIs, JSON, command line, Git.
• 动手：
  • Use curl to call a public API (e.g., GitHub API's public repo info) and extract one field from the JSON;
  • Create a local Git repo, make one commit, one change, and one rollback;
  • Read a real error log and write out "possible causes + next verification steps."
• Deliverable: records of three commands + one log analysis.

Week 4: Tech foundations II

• Read the second half of Chapter 4: data, SQL, cloud and deployment, LLMs, RAG, tool calling, security.
• 动手：
  • Create two tables in SQLite or any database, write 3 queries with JOIN and GROUP BY;
  • Draw a request chain: "user question → retrieval → model → answer";
  • Take the "20 must-know actions" self-test in Chapter 4 and mark what you don't know.
• Deliverable: tech self-test checklist + one request chain diagram.

Phase Two: Methods and Lab (Weeks 5-8) ​

Goal: Complete one closed loop from a business problem to a demoable prototype.

Week 5: Business and data

• Read Chapter 5.
• Find someone (a colleague, friend, or family member) for a 30-minute interview on the topic "the most repetitive, most annoying thing in your daily work."
• Map that role's process: inputs, outputs, roles, systems, exceptions.
• Deliverable: interview notes + one process diagram + one hypothesis of "something possibly worth automating with AI."

Week 6: Delivery methods

• Read Chapter 6.
• Using the week 5 scenario, write a 12-week deployment plan: phases, deliverables, exit criteria, risks.
• Do a "Go/No-Go" self-assessment: if data can't be obtained, the owner won't invest, or metrics can't be quantified, should you pause or continue?
• Deliverable: project charter v1 (you can use the Appendix B template directly).

Week 7: Lab A kickoff

• Read Chapter 7 and Lab A.
• Define success metrics: classification accuracy, escalation rate, human correction rate, processing time.
• Create 100 synthetic tickets (don't use real customer data), build a rules/keyword baseline.
• Deliverable: baseline results + eval set v1 (at least 30 items).

Week 8: Lab A prototype

• Connect an LLM (any OpenAI-compatible API works), build classification + confidence scoring + human review.
• Don't aim for "fully automatic." Start with a semi-automatic flow of "AI suggests + human confirms."
• Deliverables: a demo-ready prototype + a one-page architecture diagram + one failure case analysis.

Phase 3: Production and Portfolio (Weeks 9-12) ​

Goal: turn the Lab into a presentable portfolio piece, and turn the methods into a capability you can articulate.

Week 9: Evals and Production

• Read Chapter 8.
• Expand the eval set to 30-50 cases, adding edge cases: ambiguous tickets, multiple intents, PII content, malicious inputs.
• Run one regression: change a version of the prompt and compare metrics before and after.
• Deliverable: an eval report (metrics + failure cases + next steps).

Week 10: Portfolio

• Read the portfolio section of Chapter 9.
• Write Lab A up as a one-page case study: problem, baseline, solution, metrics, evidence, reflection.
• Record a 5-minute walkthrough video. Requirements: no jargon dumping—talk business first, then technology.
• Deliverables: portfolio v1 + video.

Week 11: Interviews and Communication

• Read the 8 types of interview questions in Chapter 9.
• Pick 2 questions from each type, write out your answer frameworks, and record 3-minute audio versions.
• Find someone to do a 45-minute mock interview, focusing on "problem decomposition" and "failure retrospectives."
• Deliverables: answer cards + a retrospective of one mock interview.

Week 12: Retrospective and Applications

• Read Chapter 10 (if you are in China) and the case library.
• Pick 10 target companies, research their FDE/AI delivery roles, and write tailored openers.
• Apply/get referrals, record every piece of feedback; add frequently asked questions to your answer cards.
• Deliverables: a job search checklist + one retrospective.

Differences among the three routes ​

Route A: Business to FDE (never written code) ​

Your strengths are understanding processes, people, and metrics; your weakness is engineering. Don't try to learn all of computer science first—follow the "just enough" principle:

• First learn Python basics + HTTP + SQL so you can read and modify code;
• Use low-code/off-the-shelf APIs for prototypes; don't write frameworks from scratch;
• Find a development partner: you handle problem definition, data, evals, and client communication; they handle engineering implementation;
• Fill in containers, CI/CD, and observability after 6 months.

Shortest path: Chapters 1, 3, 5, 6 → the "rules baseline + evals" part of Lab A → portfolio. You don't need to build the whole system yourself, but you must be able to explain the business rationale behind every technical choice.

Route B: Development to FDE (no client project experience) ​

Your strength is engineering; your weaknesses are business, communication, and "ambiguity." Your most likely mistake: jumping straight into code, or proving yourself through technical complexity.

Shortest path: Chapter 1 → Chapters 5, 6 → the "discovery + definition" part of Lab A → Chapter 9. Focus on practicing three things:

• Before writing any code, write out the problem statement and success metrics;
• Explain your solution to non-technical users without words like "vectors" or "fine-tuning";
• Accept that "this problem isn't worth solving with AI" is also a good conclusion.

Route C: Consulting to FDE (knows clients, shallow on tech) ​

Your strengths are interviewing, structuring, and reporting; your weaknesses are "delivery" and "technical judgment." Your most likely mistake: turning the FDE role into "on-site consulting," producing piles of PPTs but never shipping a system.

Shortest path: Chapter 4 → Chapter 6 → Lab A → Chapter 8. Focus on practicing three things:

• Personally get an API working and run an eval end to end;
• Turn "recommendations" into "runnable processes";
• Learn to say "I can't commit to that" instead of agreeing first and figuring it out later.

Four milestones ​

Don't measure progress by "chapters read"—measure it by verifiable outputs:

• M1 (end of Week 4) Concepts: without looking anything up, explain in 3 minutes how FDE differs from consulting/outsourcing, with one example of each.
• M2 (end of Week 6) Methods: for an unfamiliar scenario, write out a problem statement, success metrics, data requirements, and a 12-week plan.
• M3 (end of Week 10) Delivery: have a project that can be demoed, has evals, and includes failure analysis.
• M4 (end of Week 12) Communication: explain the project clearly in 5 minutes and answer 5 follow-up questions.

If you miss a milestone, don't rush to send out resumes. Go back to the relevant chapters to fill the gaps, or deepen the project further.

If you only have 30 days ​

Compress the 12 weeks into a 4-week sprint, keeping only the minimal closed loop:

30 days won't make you a senior FDE, but it's enough to help you judge "whether I want to keep going down this path" and give you a presentable starting point.

Things not to do ​

• Don't finish learning machine learning before starting. Most FDE work doesn't require training models—it requires knowing how to call APIs, run evals, and debug.
• Don't hoard 100 articles. Pick 1 book, 1 Lab, 1 real scenario—finishing matters more than reading everything.
• Don't build a demo with fake data and stop there. Real data will expose 80% of the problems.
• Don't only talk to IT. The real users, process owners, and economic buyers are often not in the IT department.
• Don't treat "going live" as the finish line. Launch is just the beginning—adoption, retention, and business outcomes are the evidence.
• Don't grind alone. Find a partner for code review, mock interviews, and project retrospectives—it's far more efficient.

Exercises ​

• Spend 10 minutes on the skills self-assessment, recording your scores on the four dimensions.
• Based on your background, pick a route and write down a fixed study time slot for each of the next 4 weeks (specific to day of the week and time).
• Find a study buddy and agree on a weekly 30-minute retrospective: what you did this week, where you got stuck, and what to do next week.