The write up is the part they read
Here is something that surprises people: most of the readers who matter will never run your project. They will read what you wrote about it, look at a screenshot or two, and form a judgement in a couple of minutes.
That is not laziness on their part. It is the only practical option when you are one of many applicants. Which means the write up is not a wrapper around the work. For most readers, it is the work.
It is also the only part a stranger can verify. They cannot see the four evenings you spent. They can see whether your account of what happened is specific, coherent and honest, and those three qualities are surprisingly hard to fake.
In plain English
- Write up:
- A short written account of what you built, why, and what happened.
- Attribution:
- Being clear about which parts you did and which parts a tool produced.
- Reproducible:
- Described clearly enough that someone else could follow your steps.
- Overclaiming:
- Describing work in a way that implies more than you actually did.
The five sections
Keep it to one page. Longer is not better and almost nobody reads to the bottom.
The problem. Who was losing time or making mistakes, and what it cost them. Two or three sentences. Concrete beats general every time.
What you built. What it takes in, what it produces, and how someone would use it. Plain language, no tool names in the first sentence.
What you tried that did not work. This is the section people delete and it is the most valuable one. Two things you attempted, why each failed, and what you did instead.
What you would do differently. Shows you can look at your own work critically. One or two things, specific.
Where AI helped and where you decided. Covered properly below, because getting this wrong causes real damage.
The failures section is what separates a write up from an advert. Anyone can list what worked. Being precise about what did not is evidence that you were actually there.
Be exact about what you did and what the tool did
Never present AI-generated work as entirely your own. Not because someone will catch you, though they well might, but because the interview will go somewhere you cannot follow.
Being precise about this is not a confession. It is a demonstration of judgement, which is exactly what an employer is trying to assess. The interesting question is never whether you used AI. Everyone does. The interesting question is what you decided.
โ Weak prompt
Prompt
Write a portfolio description of this project that makes me sound as capable as possible.
Output
I designed and developed an intelligent automation solution that transformed the scheduling process, leveraging advanced AI to deliver significant efficiency gains across the team.
Designed and developed implies you built it. Significant gains is unmeasured. Nothing here can be checked, and one specific question about how it works will collapse the whole paragraph.
โ Good prompt
Prompt
Write a plain description of this project in under 150 words. Separate clearly what I decided from what the AI tool generated. Include one thing that did not work. Use no adjectives about me and no words like leveraged or transformed. Here is what happened: [your notes].
Output
Drafting the weekly rota took about two hours. I wrote the prompt and the rules about which shifts cannot be adjacent, and the tool generated the draft. I decided a human still checks every draft, because it twice placed the same person on consecutive nights. Time is now roughly twenty minutes.
The reader knows exactly what you contributed. The failure and the resulting decision are the strongest lines in it, and every sentence survives a follow-up question.
Look at what the good version does. It names a rule you wrote, a specific failure, and a decision you made because of that failure. That chain is what capability looks like on paper.
Checkpoint
Write one page: the problem, what you built, what failed, what you would change, and precisely where AI helped versus where you decided.
Getting the first draft out
Writing about your own work is harder than doing it. Use your notes and let a tool interview you rather than write for you.
I need to write up a project for a portfolio. Do not write anything yet.
Interview me first. Ask one question at a time, about ten in total, covering:
- who had the problem and what it cost them
- what I actually built and what it takes in and produces
- what I tried that failed, and what I did instead
- which specific decisions I made, and what the AI tool generated
- what is still manual or still broken
After the last answer, give me a one page write up in plain language using only what I told you. Do not add anything I did not say, and do not use flattering adjectives about me.
Start with question one.
Then check the draft for the thing you cannot see in your own writing.
Here is a draft write up of my project. Go through it sentence by sentence and mark each one as:
CHECKABLE: a reader could verify this by looking at the work
VAGUE: sounds good but cannot be checked
OVERCLAIM: implies I did more than the text actually supports
For every VAGUE or OVERCLAIM, suggest a plainer replacement or tell me to delete it.
[paste your draft]
Screenshots, before, and after
Include two or three images: the input, the output, and one thing going wrong. That third one is unusual and it is memorable. It also proves you tested, which most portfolio pieces quietly fail to demonstrate.
Give a before and after in whatever units are honest. Two hours to twenty minutes. Forty errors a month to a handful. If you never measured, say you never measured, and describe the change qualitatively instead. An unmeasured claim dressed up in numbers is the fastest way to lose a reader who knows the domain.
Do not invent numbers to make the result look better, and do not round a rough estimate into a precise-sounding figure. If someone asks how you measured it and you have no answer, the entire write up becomes suspect, including the true parts.
One last pass
Read it as a stranger. Better, ask a tool to read it as one.
You are a hiring manager who has never met me and has two minutes. Read this project write up and answer:
- What can this person actually do, based only on what is written here?
- What is still unclear?
- Which single sentence would you challenge first in an interview, and what would you ask?
- Is there anything here that reads as inflated?
Be blunt.
[paste your write up]
Whatever they say they would challenge first, make sure you can answer it. That is the next lesson.
๐ Quiz
Question 1 of 4Why does the write up often matter more than the project itself?