An AI Project With No Coding Experience: Julia Started at Zero and Went to Duke
An AI project with no coding experience sounds like a contradiction, and it is the most common reason families rule themselves out. Julia Kessman had never written a line of Python when she started. She finished with a working build, and she is now at Duke studying Statistical Science and Computer Science.
That sentence is the reason this post exists, because almost every family we speak to has already decided the opposite is required. They assume a student needs to arrive as a programmer. Julia is the counterexample, and she is not a rare one.
Where an AI project with no coding experience actually starts
Not with a passion for machine learning. Not with a GitHub account. Julia came in curious and completely untrained, which put her in the same position as most students who ask us about the Intensive.
The first weeks were foundations, because that is where she was. A student who already codes starts somewhere else entirely. This is the part families underestimate: the plan is written for the student, so nobody sits through a week they do not need.
What she built
Julia worked on a HealthTech AI application built on Stanford Health imaging data, guided by Juliana Shihadeh, a published researcher whose work on bias in AI systems appeared in AI & SOCIETY.
One piece of honesty about this project. At the time, our STEAM in AI cohorts ran a shared build alongside individual work, so Julia and another student, Siyona Agarwal, worked on the same underlying project. We have since moved to individually designed projects, and we would flag identical cohort projects as a warning sign if you saw them elsewhere today. We say so plainly in our own piece on whether a $5,000 AI program is worth it.
It is worth telling anyway, because of what happened next.
One AI project with no coding experience, two different destinations
Julia took that work toward Duke, for Statistical Science and Computer Science. Siyona took the same underlying build toward UC Davis, for Biomedical Engineering.
Same data. Same problem. Two students who framed it around what they cared about, emphasised different parts of it, and ended up in majors that share almost nothing.
This is the thing we most want families to understand. A project is not a major. The project is raw material, and what a student does with it, which questions they chase and which parts they can talk about for twenty minutes, is what makes it point somewhere.
If your student is holding back because they cannot code, they are holding back for the wrong reason.
Julia would not have applied under her own assumptions about herself. We admit on a fit assessment rather than a coding test, and the assessment is free, so the only cost of finding out is a conversation.
How STEAM in AI built a project for someone who had never coded
Julia was not handed a beginner track. Before her twelve weeks began, her STEAM in AI project mentor wrote a week-by-week schedule for her specifically, starting where she actually was and ending at the project she actually wanted. Week one was Python. It is written that way for every student, which is why a beginner and a fluent coder can sit in the same cohort without either one being in the wrong room.
She also had two mentors rather than one: a curriculum mentor across the cohort, and a separate 1:1 project mentor matched to her project. That is the structural reason a first-time coder finished a working build instead of a tutorial.
The same project, five other ways
Medical imaging sounds narrow. It is not, and this is the exercise we run with families who cannot yet see how a project connects to their student’s direction.
| Toward economics or public policy | The same model, pointed at access. Who gets scanned, who does not, and what a diagnostic tool costs a health system per case avoided. |
|---|---|
| Toward pre-med or public health | The clinical side. What false negatives mean in practice, and where a model should defer to a physician instead of producing a number. |
| Toward design or human-computer interaction | The interface. How a radiologist should see a confidence score at two in the morning without being misled by it. |
| Toward law or ethics | The consent and liability questions. Whose data trained it, who is accountable when it is wrong, and what disclosure should look like. |
| Toward business | The adoption problem. Why hospitals do not buy proven tools, and what the procurement obstacle actually is. |
Five students, five different applications, one underlying build. None of these is a stretch, and none of them requires the student to want a career in medicine.
We now design each student’s project individually from the start, which means this reframing happens before the work begins rather than after. What a 12-week plan looks like when it is written for one student is covered in our week-by-week walkthrough.
Why an AI project with no coding experience worked for her
Not the topic, and not the tooling. An AI project with no coding experience behind it still has to survive the same test as any other. Julia could explain what she had done and why, which is the whole test. She made real decisions, hit real problems, and could account for both afterwards.
That has become measurably more important since. Penn now asks applicants whether they could explain, in conversation, how they arrived at what they wrote, and other universities have quietly reduced essay requirements over AI concerns. We covered both in Penn’s new rule on AI in college applications.
A student who built something and struggled with it passes that test without preparing for it. A student handed a finished template does not.
What this means for your student
Julia’s story is not useful because it ended at Duke. It is useful because of where it started, which was nowhere in particular.
Her sister Alexandra joined afterwards, having watched the whole thing up close, which is the endorsement we take most seriously. Families who have seen the process from the inside send us their second child.
FAQ: Julia Kessman’s story
Did Julia really have no coding experience?
None. She had never written Python before she started, and the early weeks of her plan reflected that.
Does an AI project with no coding experience have to be HealthTech?
No, and that matters. Julia’s cohort ran a shared build alongside individual work. Projects are now designed individually from the start, so no two students in a cohort produce the same thing.
How can the same project suit two different majors?
Because the framing does the work. Julia emphasised the statistical questions and Siyona the engineering ones. Same build, different centre of gravity, different application.
Who mentored her?
Juliana Shihadeh, a published AI-bias researcher whose work appeared in AI & SOCIETY. Students meet their mentor before enrolling rather than after. Others are introduced in Meet the mentors.
How should a student begin an AI project with no coding experience?
Find the problem before the tool. The students who finish are the ones who started from something that genuinely bothered them. More in AI programs for students who are not coders yet.
Is a project like this enough on its own for a school like Duke?
No, and nobody should suggest otherwise. Julia had a full application. The project gave her something specific and defensible to write and speak about, which is a real advantage and not a guarantee.
Every cohort has a Julia in it, and she never thinks she qualifies.
The students who arrive certain they belong are not usually the ones who go furthest. If your student has an interest and no idea what to do with it, that is the profile we are actually looking for, and cohorts are small because each plan is written individually.
