# Build an FDE Portfolio That Proves You Can Ship for a Customer

Author: Sachin Bhatnagar | Published: 2026-11-19 | Canonical: https://stacksauce.dev/blog/build-an-fde-portfolio-that-proves-you-can-ship-for-a-customer

## **Build an FDE Portfolio That Proves You Can Ship for a Customer**

In Part 1, I explained what a Forward Deployed Engineer does. An FDE sits close to the customer and makes a product work for that customer's real problem. In Part 2, I gave you six layers to learn in order, and one project that grows with you through all six.

This part is about showing that work to someone who might hire you. A portfolio is a small collection of your projects that proves what you can do. For most developer jobs, a portfolio shows that you can build. For an FDE job, it has to show something harder. It has to show that you can build for somebody else, in their messy world, and hand it over in a way they can use.

So let me show you how I would build one, step by step.

### **What FDE hiring managers are looking for**

Start with the job post itself. OpenAI's posting for a Forward Deployed Software Engineer asks people to "embed deeply with strategic customers" and to code "side-by-side to drive projects to completion on their infrastructure". It also asks them to prepare "detailed scopes of work and project plans" and to contribute to internal knowledge bases with what they learnt from customers.

Read those lines again slowly. Understanding the customer comes first, then planning, then building on the customer's own systems, then writing it all down for others. Coding is one part of a longer chain. Many learners hide one good project under a pile of half-finished clones, and the good one never gets seen.

One honest note before we go on. That OpenAI posting asks for 7 or more years of full stack experience. Senior FDE roles are like that. Your first job may be a junior FDE role, a support engineering role or a solutions engineering role. The portfolio below works for all of them, because all of them reward the same thing: proof that you can solve a real person's problem.

### **The idea: invent one customer and serve them properly**

You do not have customers yet. That is fine. Invent one, and treat them as real.

Pick a small business you understand well. A coaching centre, a medical store, a hardware shop, a wedding caterer or a school transport service. If someone in your family runs a business, even better. 

A few weeks ago, I met a friend whose father is a Chartered Accountant. When our conversation turned to AI, he immediately asked whether it could help streamline his father's work, particularly since staff occasionally make avoidable mistakes.

I knew a few students who were learning AI, so I connected them with my friend. The last time I checked, they had proposed a prototype built around multiple AI agents that could ingest, categorise, and index documents. The system would allow my friend's father to retrieve documents using natural-language queries and build processing workflows from there.

It's still an MVP, but the potential is clear. Plus, the students got the perfect project to showcase their Agentic AI skills.

Now write down their problem the way they would say it. For example: "Every month I spend two days matching bank payments with student fee records in Excel, and I still miss some." That one sentence is worth more than any fancy project idea, because a real person could have said it.

Your whole portfolio grows from that one customer. The six layers from Part 2 become the six things you did for them.

### **The five pieces of an FDE portfolio**

#### **1. A one-page problem brief**

Before any code, write one page. Who is the customer? How do they handle it at the moment? What goes wrong? What does "fixed" look like to them? Keep it in plain words, with no technical terms.

This page copies the "scope of work" that FDE job posts mention. A scope of work is a short agreement on what you will build, what you will leave out and how everyone will know it is done. Most beginners skip this step. Hiring managers notice when you don't.

#### **2. Messy data, cleaned with care**

Real customers never hand you clean data. So make your sample data messy on purpose, or find a public dataset that already is. Use dates in two formats, names with spelling mistakes, a few empty cells and one duplicate row.

Then clean it with Python and SQL, and write down every decision you made. "I treated two rows with the same phone number as the same student" is the kind of line that shows judgement. Layer 2 from Part 2 lives here.

#### **3. A deployed tool with a simple screen**

Deployed means it runs on the internet, so anyone with the link can use it. A project that only runs on your own laptop is very hard for a hiring manager to judge.

Give your customer a simple page. A form, a table and one or two buttons are enough. Connect it to at least one outside service through an API, which is a set of rules that lets two programs talk to each other. A payment sheet, a messaging service or a spreadsheet all work. If it makes sense, add one AI feature on top, like a summary of the month's missing payments.

Keep the scope small. A narrow tool that works every time is far better than a big one that works only when you are watching.

#### **4. A debugging story**

This is my favourite piece, and the one most learners leave out. Something will break while you build, and that is good news for your portfolio. Write it up.

Use five short headings. What should have happened? What happened? How did you find the cause? What did you change? What would you do differently next time? One or two pages is plenty. Be honest about the fixes that failed before the one that worked.

Why does this matter so much? Because an FDE spends a large part of the job inside other people's broken systems. A calm, clear debugging story shows a hiring manager how you think under pressure, and that is very hard to fake.

#### **5. A handover note for a non-technical reader**

In Part 2, I asked you to write a one-page note after every project. Here it becomes the final piece. Imagine the coaching centre owner has to use your tool after you leave. What do they need to know? How do they log in, what do the buttons do, and who do they call when something looks wrong?

Show this note to someone who does not code. If they can use the tool from the note alone, you have written a good one.

### **How to lay it out so it can be read in two minutes**

A hiring manager may give your portfolio two minutes on a busy day. So put everything on one page, in this order.

1. **One line on the customer and the problem.** For example: "A coaching centre that loses two days a month matching fee payments."
2. **One line on the result.** For example: "A web tool that matches payments in under a minute and flags the ones it cannot match."
3. **The live link and a short screen recording.** Ninety seconds is enough. Talk the viewer through it as you would talk to the customer.
4. **Links to the brief, the debugging story and the handover note.**
5. **The code repository last.** A repository is the online folder, usually on GitHub, that holds your code and its history.

Notice the order. The customer comes first and the code comes last. Someone who reads only the first two lines should already know what you did and for whom.

### **"But Sir, a fake customer is still fake"**

Now most will say, "Sir, the interviewer will know my customer is imaginary. What is the point?" Partly agreed. A pretend customer is weaker proof than a real one. If you can find a real one, do it. Many small businesses would happily let a student fix one annoying problem for free, and their words in your brief carry real weight.

But even a pretend customer forces you to think in the right order. You start from a problem, you make choices about data, you build something someone else can use and you write for a reader who does not code. That way of thinking is exactly what an FDE interview will test in Part 4. The habits you build while serving that customer stay with you, whoever the customer was.

### **Mistakes I would avoid**

- **Tutorial clones.** If a thousand people built the same app from the same video, yours tells the interviewer very little about you.
- **A long list of tools with nothing behind them.** Every skill on your CV should point to a place in your project where you used it.
- **Only screenshots.** Give a live link. If the free hosting goes to sleep, say so on the page, so the reviewer waits a few seconds.
- **Hiding the failures.** A project with no problems in its story looks like a project nobody really tested.
- **Too much AI on top.** One useful AI feature that you have tested is plenty. Five shiny ones that sometimes give wrong answers will worry a careful reviewer.

### **What this means for you**

1. **Pick your customer first.** Write their problem in one sentence, in their words.
2. **Write the one-page brief before any code.** Keep it in plain language.
3. **Keep building your six-layer project from Part 2,** but for this one customer.
4. **Start a "things that broke" file on day one.** Each entry takes two minutes, and it becomes your debugging story later.
5. **Run the two-minute test.** Send your portfolio page to a friend. Give them two minutes. Ask them who the customer was, what problem you solved and whether it works. If they cannot answer all three, fix the top of the page first.

One portfolio done this way will carry you further than ten done the usual way. It takes longer, I know. That extra time is also why very few applicants will have a portfolio like yours.
