What Should Be in an AI Project Portfolio for Students?
One deployed project per type of role you apply to, each built from a real job description for that role. Here is what belongs in each entry and what does not.
By Howard Davner
💡 Quick Answer One deployed project per type of role you apply to, each built from a real job description for that role, each reachable at a URL that opens in a browser. That is the whole portfolio. Not ten tutorial repos, not a list of coursework, not a GitHub profile. The count is small on purpose: the test is fit, not volume.
Most portfolio advice tells students to build more. The advice is wrong in a specific way: it treats a portfolio as evidence of effort, when the person reading it is using it as evidence of fit.
Those two readings want opposite things. Evidence of effort wants quantity. Evidence of fit wants one thing that clearly belongs next to this job.
Here is what that means in practice.
One Project Per Role Type, Not One Per Application
You need one project for each kind of role you are applying to, not one for each application you send. A student applying to data science internships and software engineering internships needs two projects, because those are two different briefs. That same student applying to thirty data science internships still needs one.
The distinction matters because it changes the size of the job from impossible to a weekend. "Tailor your portfolio to every application" is advice nobody follows, because nobody can. Tailoring it to every kind of application is two or three projects, total.
If you want the longer argument about counts, how many projects you actually need on a resume works through it with the resume in view.
Why the Job Description Is the Only Brief That Matters
Build the project from a real job posting, because that posting is the one document you and the recruiter both have open. Everything else in your application is your account of yourself. A project built against their own requirements list is the only item in the pile that can be checked line by line against something they wrote.
That is the whole mechanism, and it is why a generic project loses to a specific one even when the generic project is technically harder. A weather dashboard is an accomplishment with no addressee. A tool that does the thing the third bullet of the posting describes is addressed to the person reading it.
The practical version: open a posting for a role you actually want, find the sentence that describes a task rather than a qualification, and build the smallest working thing that does that task. Provieo's project builder does this from a pasted job description, matching the posting to a project scoped small enough to finish and then walking the build through to a deployed URL.
What Each Portfolio Entry Actually Contains
Four things, and the entry is not finished until it has all four:
- The live URL. Written out in full, not hidden behind the words "my portfolio". This is the only mandatory element.
- One sentence naming the problem. Not what the project is built with. What it does for whom.
- The stack, in one line. Enough for a technical reader to know what they are looking at.
- One decision and why. The thing you chose to do differently, or the thing you deliberately left out. This is the line interviews get built on.
What is not on the list: a screenshot gallery, a long write up, a logo, or a count of hours spent. None of those survive a reader who is thirty seconds into your application.
Do You Need Machine Learning Skills to Build One?
No. Most strong student AI projects call an existing model API rather than training a model, and that is not a shortcut, it is the same thing the teams hiring you do. The skill on display is scoping a real problem, wiring a usable tool around a model, and shipping it somewhere a stranger can open it.
Training a model from scratch demonstrates a genuinely different skill, and most entry level roles are not hiring for it. If you are weighing a course against a build, the certificate versus project comparison sets out what each one actually proves.
How a Recruiter Checks the Work Is Yours
Three ways, and none of them involve reading your code. They open the URL and use the thing. They glance at whether the commit history shows work spread across days rather than a single upload. And they ask, in the interview, why you made one particular choice.
That third one is the reason the "one decision and why" line matters. A project you built has an answer to it, immediately and in your own words. A project you copied has a pause. This is also why volume is a bad strategy: ten projects you cannot each talk about in detail is a worse position than one you can.
For more on what the hiring side is actually scanning for, what employers look for in a CS portfolio is written from that side of the desk.
What It Costs to Host, and How Long One Takes
Hosting a small student project can cost nothing. Vercel and Hugging Face Spaces both have free tiers that will serve a project at a public URL. Check the current limits before you depend on one, and check the link again the week you start applying, because a portfolio URL that returns an error is worse than no URL at all.
Time is the wrong thing to plan. Scope is the thing to plan. Decide what the first working version does not do, get that version deployed, and improve it afterwards if you want to. A project scoped to one task and finished beats a project scoped to five and abandoned, and the second kind is what fills most student GitHub accounts. Building and deploying in one focused sitting walks the fast path.
Keep Building: Portfolio Guides
This page is the hub. Each of these goes deeper on one piece of it:
- AI project ideas for your resume, role by role
- AI portfolio projects worth deploying, with build guides and stacks
- Why a deployed project on your resume changes the conversation
- Does GitHub count as a portfolio?, on why a live URL beats a repository
- What a deployed project needs for a recruiter to trust it
- CS portfolio projects for internships
- Portfolio versus resume for tech jobs, on how the two work together
Provieo is free and unlimited to use. Your first project build and deploy is free, and after that it is one flat one-time payment for unlimited deployed projects forever, with no subscription.
Frequently Asked Questions
What should be in an AI project portfolio for students?+
One deployed project per type of role you apply to, each built from a real job description for that role, each reachable at a URL that opens in a browser. Each entry needs four things: the live URL, one sentence naming the problem it solves, the stack it runs on, and one decision you made and why. A GitHub profile, a list of coursework, or a personal site describing projects without linking to running ones is not a portfolio.
How many AI projects does a student portfolio need?+
One per type of role, not one per application. A student applying to data science internships and software engineering internships needs two, because those are two different briefs. Applying to thirty data science internships does not require thirty projects. Applying to a different kind of job does.
Do you need machine learning skills to build an AI portfolio project?+
No. Most student AI projects call an existing model API rather than training a model. The skill being demonstrated is scoping a real problem, wiring a working tool around a model, and shipping it somewhere a stranger can open it. Training a model from scratch demonstrates a different skill that most entry level roles are not hiring for.
Does an AI project portfolio count if you are not a computer science major?+
Yes, and the project should match your field rather than imitate a CS one. A marketing student building a tool that drafts and scores campaign copy is showing exactly the judgment a marketing team hires for. The rule is the same in every field: build the thing the role would actually use, deploy it, link it.
How does a recruiter check that a portfolio project is really yours?+
Three ways, and none of them require reading your code. They open the live URL and use it. They look at whether the commit history shows work spread over time rather than one upload. And they ask you in the interview why you made a specific choice, which is a question you can only answer about a project you actually built.
Where can a student deploy an AI project for free?+
Vercel and Hugging Face Spaces both have free tiers that will host a small student project at a public URL. Check the current limits before you rely on one, because free tiers change. What matters is that the URL stays up, because a portfolio link that returns an error is worse than no link.
Is a GitHub profile enough on its own?+
No. A repository is source code, which requires the reader to clone it, install it and run it before it shows them anything. A deployed project shows them something the moment they click. Keep the repository and link to it from the project, but the URL in the portfolio field should be the running thing.
Put a working link in your next application.
Paste the job you're applying to. Provieo designs a project against that exact listing and walks you through deploying it, so what a recruiter clicks is the work itself, not a description of it.
Free to use, and your first deployed project is free. Or tailor your resume to the same posting.