An AI portfolio project an employer can check is small, real and open. It solves one narrow problem, shows the input, your prompts, the output and how you checked it, and lives at a link anyone can open. It also says honestly what did not work. That is far more convincing than “proficient in ChatGPT” on a CV.
This article walks you through choosing, building and presenting one such project. It is written for students and freshers from any stream, including commerce, arts and engineering. You do not need to write code, although you can. It does not promise that a project will get you a job; nothing can. It will give you something concrete to show and talk about.
Why does a list of AI tools on a CV not convince employers?
Because it is a claim that cannot be checked. A list of tools, a screenshot of a chat or a certificate of attendance tells an interviewer very little about what you can actually do.
From an employer’s side, three questions matter: can you use AI to do useful work, can you tell when it is wrong, and can you explain your choices. A portfolio project should answer all three, in a form they can open in two minutes.
What makes a project “checkable”?
A checkable project is one where someone else can see your inputs, follow your steps and judge the result for themselves. If they have to take your word for it, it is not checkable.
In practice, that means:
- A public link. A shared document, a simple web page or a public code repository. Not a file on your laptop.
- The real inputs. The sample text, data or brief you started with (anonymised or made up, never real personal data).
- Your prompts. The actual prompts you used, including the ones that did not work.
- Your checks. How you verified the output, and what you corrected.
- A before and after. What the task looked like without AI, and with it.
How do you choose the right project?
Pick a narrow, real task from the kind of job you want. Narrow is the important word: one task done well and documented clearly beats an ambitious project you cannot finish.
Some ideas by background, as starting points:
| If you are aiming for | A narrow project idea |
|---|---|
| Marketing or sales | A repeatable process for turning one product description into three channel-specific posts, with a checklist for brand tone and factual accuracy |
| Operations or admin | A template that turns rough meeting notes into action items with owners, tested on sample notes you wrote yourself |
| Finance or commerce | A workflow that summarises a public annual report into a one-page brief, with every figure checked against the original |
| HR or people roles | A structured prompt that drafts a job description from a hiring manager’s notes, checked for biased or exclusionary wording |
| Engineering or IT | A small script or tool that uses an AI model for one task, with a test set showing where it works and where it fails |
Use public material, or sample material you create yourself. Do not use data from an internship or employer without written permission, and never use real customer or student personal data.
What should the project write-up contain?
Six short sections, in this order. Keep the whole write-up readable in five minutes.
- The problem. One paragraph. What task, for whom, and why it matters.
- The starting point. How the task is done without AI, and how long or how hard that is, in your own honest estimate.
- The approach. Which tool you used (for example ChatGPT, Gemini, Claude or Copilot), your prompts, and how you refined them. Show at least one prompt that did not work and what you changed.
- The checks. How you tested the output. For example: you ran it on several sample inputs and marked every error, or you compared figures against the source.
- The result. A before-and-after example, side by side.
- The limits. Where the approach fails, what it should not be used for, and what you would do next with more time.
Do not skip the limits section. It shows judgement: that you know when not to trust the tool.
How do you show your checking, not just your output?
Keep a simple log as you work: input, prompt, output, what was wrong, what you changed. Then include a short version of it in the write-up.
For example, if you built a meeting-notes template, test it on several sets of sample notes you wrote, some messy on purpose. Record where it missed an action item or assigned the wrong owner, and how you changed the prompt. A small table of “test, problem found, fix” is enough.
This is also the habit that matters most on the job. My routine for it is in Checking AI output before you send it, and the prompt structure I recommend is in A simple prompt structure for reports and emails.
Try it: Write down three tasks you did in college or an internship this month that took more than an hour. Circle the narrowest one. That is your first candidate project.
How do you present the project in an interview?
Lead with the problem, not the tool. Then walk through one example, one mistake you caught and one limit you found.
A simple structure for a two-minute answer:
- “The task was…” (one sentence on the problem).
- “Here is what I tried first, and why it did not work.”
- “Here is what I changed, and how I checked the result.”
- “Here is where it still fails, and what I would do next.”
Have the link ready to share on screen. If the interviewer asks to try it, let them. A project that survives a stranger’s test is a strong signal.
Where do certifications fit alongside a project?
A certification and a project show different things. A certification shows you passed an independent assessment; a project shows how you work. They are stronger together than either is alone.
If you hold an IBAIC credential, list it with its credential ID. Anyone can check an IBAIC credential by its ID on the IBAIC public register, which makes it checkable in the same way as your project. If you are deciding whether to start with L1, read IBAIC L1 vs L2: which one should you sit first?.
What should you avoid?
Avoid anything an interviewer could not verify or would not trust.
- Inflated claims. “Saves hours every week” means nothing unless you show how you measured it. If you did not measure it, do not claim it.
- Real personal or confidential data. Use made-up or public data only.
- Copied projects. Following a tutorial step by step is learning, not a portfolio piece. Change the problem and make it yours.
- Hiding the AI. Be clear about what the tool did and what you did. That is the point of the project.
What to do next
Choose one narrow task this week and start your log with the first prompt you try. If you want a structured path with guided projects and IBAIC L1 preparation, see the students path on the learn page.