United States · Grades 10 to 12 · Machine learning research

AI research programs for high school students

Plenty of teenagers now train a model. Far fewer turn one into research, which means asking a question nobody has answered, testing it fairly, and writing up what happened, including what failed. American high schoolers have a handful of serious places where that kind of work is read and judged. This guide takes the ones where machine learning fits naturally, quotes their current rules, and is plain about the new part: what each organizer lets a student do with generative AI, and what it does not.

Book the free first class Book a priority demo Read the STS AI rules Ask on WhatsApp

Live teaching since 2020 · 10,000+ students · Research skills taught, never research written for you

The short answer

The main US places where a high schooler's AI research is judged are Regeneron STS (seniors, individual projects, deadline November 5, 2026), Regeneron ISEF through a Society-affiliated fair (category Robotics and Intelligent Machines, subcategory Machine Learning), and summer research at RSI or BU RISE for juniors. Modern Age Coders teaches the method, Python and statistics live online, ages 13 to 18, in a batch or one to one.

Build the research toolkit

Three courses behind a credible AI research project

Research needs a model, a dataset handled honestly, and the reasoning to defend a result. Each course is taught live by a regular teacher.

AI and Machine Learning for Teens course thumbnail

ARPH / 01

AI and machine learning for teens

Ages 13 to 18. Models built from the inside, so results can be explained.

Open the syllabus →
Data Science for Teens course thumbnail

ARPH / 02

Data science for teens

Ages 14 to 18. Clean data, fair splits and charts a judge can read.

Open the syllabus →
Problem Solving for Teens course thumbnail

ARPH / 03

Problem solving for teens

Ages 13 to 18. The algorithmic thinking that turns a hunch into a method.

Open the syllabus →

Where AI research is judged

Four programs, from their 2026 to 2027 rules

These are the programs where a machine learning project fits naturally. Each row comes from the organizer's own page or rule book, read on 17 October 2026. None of them is ranked here; they do different jobs.

Research programs open to US high schoolers where AI work fits (from the organizers' pages, read 17 October 2026)
ProgramRun byWhoKindKey date
Regeneron Science Talent SearchSociety for ScienceSeniors, final year of secondary schoolIndividual research, judged on a written reportApplications due November 5, 2026, 8:00 pm ET
Regeneron ISEFSociety for ScienceHigh schoolers who win at an affiliated fairScience fair, 22 categories in 2026ISEF 2027 in Los Angeles, May 2027
Research Science InstituteCenter for Excellence in EducationHigh school juniors onlySummer: classes, then a five-week internshipApplications close 11:59 pm, December 11, 2026
BU RISEBoston UniversityJuniors, domestic studentsSix-week summer lab internship or practicum2027 practicum subjects announced in December
MIT PRIMES-USAMIT MathematicsJuniors and sophomoresFree year-long remote mentoringDeadline November 2, 2026

Competitions and programs are different things

STS and ISEF judge finished work. They do not supply a mentor or a lab; the student arrives with a project. RSI, RISE and PRIMES are the opposite: they place a student with a mentor and the research happens inside the program.

A realistic plan for an AI project aimed at STS spreads the work over a junior year and a summer, sometimes inside one of the mentored programs. The two kinds fit together rather than compete.

What fits an AI project, and what does not

PRIMES-USA currently describes its research projects as being in mathematics and bioinformatics, so it suits a student whose AI interest is mathematical rather than applied. Read its own page for the current list before applying.

At RISE, computer science is one of the internship areas the program lists. At RSI, each student works on an individual project with a mentor who is an experienced scientist or researcher. Both are summer programs, so their deadlines fall in the school year before.

Sources, read 17 October 2026: Regeneron STS and its 2027 rules; Regeneron ISEF; RSI at CEE; BU RISE; MIT PRIMES. We are not connected with any of these organizations. For the selective summer programs in depth, see our summer research programs guide.

Regeneron STS 2027

Regeneron STS: the eligibility rules and the generative AI rules

The 2027 rule book includes a full appendix on generative AI in student research. It is worth reading in full; here is the part every AI-minded senior needs.

Regeneron STS 2027 eligibility, from the official rules (read 17 October 2026)
RuleWhat the rules say
AgeOver 13 with parental or guardian consent, or 18 or older but under 21
School yearApplicants must be in their final year of their secondary school, and must not have entered any previous STS
Type of workApplicants must complete individual research projects; team research is not eligible
Team projectsStudents may not "split" a team project and enter as individuals
Working in a labResearch conducted alongside adult researchers in a research institution is permitted, if the student's own role is clear
The paperThe paper should be 20 pages or less, with title page, abstract and bibliography outside the limit
DeadlineApplication and recommendations by November 5, 2026 at 8:00 pm Eastern Time
Generative AI in an STS project: selected rows from Appendix 4 of the 2027 rules
TaskAllowed?Condition in the rules
Writing application answers, the Research Report or citationsNoRule 8: "Students may not use generative Artificial Intelligence (AI) to write Regeneron STS application questions, draft the Research Report or generate citations."
Using AI to write initial code for the projectYesOnly with explicit citation of which code was AI generated, and a log of the prompts
Asking AI which statistical tests or tools fitYesRequires a log of your prompts; interpreting the data stays with the student
Asking AI for a starter bibliographyNoNever acceptable
Using AI to produce conclusions or future stepsNoNever acceptable
Polishing grammar in an abstract the student wroteYesMinor grammar and syntax changes only, and credited

What this means for an AI project

The rules draw a clear line: "Use of AI tools for student research projects is permitted, and should be disclosed and appropriately cited." A model you train is a research tool. A chatbot that writes your discussion section is not allowed.

Keep a prompt log from the first week. It costs a minute a day and it is the evidence the rules ask for if any code or analysis advice came from an AI.

The numbers, read honestly

Society for Science says around 2,600 students enter each year and 40 become finalists. That is 40 / 2,600, about 1.5 in 100, our arithmetic. Entering is still worth it: the rule book itself is a short course in research ethics.

Top 300 scholars are announced on January 7, 2027 and the top 40 finalists on January 21, 2027. Finals Week runs March 11-17, 2027 in Washington, D.C.

Regeneron ISEF

Regeneron ISEF: where a machine learning project is entered

ISEF is reached through a local, regional or state fair affiliated with Society for Science. The category choice matters, because judges are assigned by subcategory.

The ISEF category that fits a typical machine learning project (our view), with definitions from the Society for Science category pages (read 17 October 2026)
ItemWhat Society for Science says
CategoryRobotics and Intelligent Machines (ROBO)
Definition"Studies in which the use of machine intelligence is paramount to reducing the reliance on human intervention."
SubcategoriesBiomechanics, Cognitive Systems, Control Theory, Machine Learning, Robot Kinematics, Other
Machine Learning (MAC)"Construction and/or study of algorithms that can learn from data."
Other homes for AI work (our suggestion)Software Design, Computational Biology and Bioinformatics, Embedded Systems, Mathematics
Category countThe following 22 categories are those that will be used at Regeneron ISEF 2026

Choosing between ROBO and Software Design

If the question is about the learning itself (does this model generalize, which features matter, how does accuracy change with data size), the Machine Learning subcategory is the natural home.

If the project is an app that happens to use a trained model, and the contribution is the software, Software Design may fit better. Society for Science asks students to choose the primary subcategory rather than Other.

Getting there

The 2026 finalists were selected from 365 affiliate fairs. Your school or district fair is the first step; ask your science teacher which fair it feeds.

Local and state fairs may use their own categories, and Society for Science says so. Check your affiliated fair's list before you print a board.

Sources: ISEF categories and subcategories and the full category definitions, read 17 October 2026. Competition is a good reason to finish a project; for event-style alternatives see hackathons for high school students and the Congressional App Challenge guide.

A worked example

The first experiment every AI research project needs: beat a baseline

Judges ask one question of every machine learning project before any other: better than what? This script is ours. It compares three versions of one model against a baseline that always guesses the most common answer, using five-fold cross-validation on the handwritten digits dataset that ships with scikit-learn. Results below are from our run of 17 October 2026 (scikit-learn 1.7.2, Python 3.13).

An honest comparison in 20 linespython
# A fair test: model versus baseline (ours)
from sklearn.datasets import load_digits
from sklearn.dummy import DummyClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import cross_val_score

X, y = load_digits(return_X_y=True)

contenders = {
    "baseline": DummyClassifier(strategy="most_frequent"),
    "knn k=1": KNeighborsClassifier(n_neighbors=1),
    "knn k=5": KNeighborsClassifier(n_neighbors=5),
    "knn k=25": KNeighborsClassifier(n_neighbors=25),
}

for name, model in contenders.items():
    scores = cross_val_score(model, X, y, cv=5)
    print(f"{name:9} mean {scores.mean():.3f}"
          f"  spread {scores.min():.3f}-{scores.max():.3f}")

Output on our run: baseline mean 0.101; knn k=1 mean 0.964 (spread 0.953 to 0.986); knn k=5 mean 0.963 (0.947 to 0.981); knn k=25 mean 0.946 (0.922 to 0.967).

Reading the result like a researcher

The digits dataset has 10 classes, so a model that always picks the most common digit should be right about 1 / 10 = 0.1 of the time. The baseline's 0.101 confirms the test is set up correctly before anything else is believed.

k=1 and k=5 differ by 0.001 in their means while their spreads overlap almost entirely. A careful write-up says they performed about the same on this data, not that k=1 won.

Turning it into a question

The script is a method, not a finding; any student can run it. Research starts when the student changes something nobody has tested in their setting: their own dataset, a deliberate imbalance, or noise added to see where the model breaks.

scikit-learn describes DummyClassifier as a simple baseline to compare against other classifiers. Using one is the cheapest way to look rigorous, because it is rigorous.

  1. Write the question in one sentence. If it needs two, it is two projects.
  2. Name the baseline before training anything. Most frequent class, a simple rule, or last year's method.
  3. Fix the split. Cross-validation or a held-out test set, chosen before you see any result.
  4. Report the spread, not just the mean. Overlapping spreads mean no clear winner.
  5. Change one thing at a time. Two changes at once make the result impossible to explain.
  6. Write down what failed. A dead end that is explained is a result, and a judge reading the report will want to see it.
  7. Log every AI assist. Prompts, dates and which code came from where, as the STS rules require.

Timing

A research calendar for an eleventh grader aiming at senior-year programs

Our suggested calendar, working back from the STS deadline of November 5, 2026 for this year's seniors and from the same point a year later for today's juniors. Dates in the third column are the organizers'; the rest is ours.

Grade 11 to grade 12, built around published deadlines (our plan; organizer dates as published)
WhenStudent work (ours)Published dates to know
Grade 11, fallLearn the tools: Python, a first model, a baselinePRIMES-USA deadline November 2, 2026; RSI 2027 closes December 11, 2026
Grade 11, winterPick a question; find or collect a datasetBU RISE announces 2027 practicum subjects in December
Grade 11, springPilot experiment; enter a local affiliated fairAffiliated fair dates vary; ISEF 2027 is in May 2027 in Los Angeles
Summer before grade 12Main experiments, in a program or at homeRSI and RISE run in summer
Grade 12, September and OctoberWrite the report yourself, under 20 pagesSTS for that year closes in early November

For tenth graders

Spend this year learning to code and to read a results table. The calendar above starts next fall, and students who arrive with the tools move faster.

For juniors now

The RSI and PRIMES deadlines above are this school year. Apply if the fit is right, and keep a home project going either way.

For seniors now

The STS deadline is November 5, 2026. If the research is done, the remaining work is the writing, and the rules say that part is yours alone.

Getting help the right way

What a teacher can and cannot do for a student researcher

Every program here expects the work to be the student's. The STS rules require entrants to disclose support of any kind, naming paid support and AI use among the examples. That shapes how we teach.

What we do

Teach the skills: Python, scikit-learn, experimental design, statistics, and how to read a paper. A student who understands cross-validation can design their own experiment.

Review the method, not write it. We ask the questions a judge would ask and the student answers them, in their own code and their own words.

What we never do

Choose the question, run the experiments, or draft any part of a report or application. That would break the rules and it would not be the student's research.

Promise a result. No teacher can promise that a project places, and we do not. A family should disclose our classes if a program asks about outside support.

If your teenager is still at the project stage, our AI projects for high school students page sets out a ladder from a first model to research-grade work. For extracurriculars beyond research, see computer science extracurriculars.

From project to research

Four stages between a tutorial and a research entry

Where an AI-curious high schooler might be, and what separates each stage from the next.

Stages of a student AI researcher (ours)
StageWhat it looks like
1. FollowerCompletes tutorials; results match the tutorial
2. BuilderTrains a model on a dataset they chose
3. TesterCompares against a baseline and reports a spread
4. ResearcherAsks a new question and writes up what failed

For parents

Stage 3 is the real turning point. Ask your teenager what their model was compared with; a clear answer means they are thinking like a researcher.

A project that does not place can still be excellent research. The write-up is what lasts.

For students

Pick a question small enough to finish. A tight question with an honest answer beats a grand one that stops halfway.

Save every result, including the bad ones. Your future report needs them.

The course catalog

Nine courses for teenagers heading toward research

Live online with a regular teacher, in a batch or one to one.

I

Foundations

Ages 13 to 18

ARPH / FND / 01

Python for teens

The language almost every student ML paper is written in.

Open the syllabus

ARPH / FND / 02

Problem solving

Breaking a question into steps a program can test.

Open the syllabus

ARPH / FND / 03

MySQL for teens

Storing and querying a dataset you collected.

Open the syllabus
II

Models and data

Ages 13 to 18

ARPH / MOD / 01

AI and ML for teens

From one neuron to networks, measured honestly.

Open the syllabus

ARPH / MOD / 02

Data science for teens

Cleaning, charting and testing a claim.

Open the syllabus

ARPH / MOD / 03

Orange data mining

Fast pilot experiments before writing code.

Open the syllabus
III

Showing the work

Teens

ARPH / SHW / 01

Hackathon preparation

Presenting a technical result under time pressure.

Open the syllabus

ARPH / SHW / 02

React for teens

A clean demo page for a finished model.

Open the syllabus

ARPH / SHW / 03

AI tools for teens (13+)

Using assistants within the rules, with a prompt log.

Open the syllabus

How lessons run

Classes that work around a research timeline

Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays.

Weekdays

An evening slot that suits your US time zone.

Weekends

Weekend mornings for longer experiment sessions.

1-on-1

Paced to a fair date or a deadline.

A steady teacher

One teacher who knows the project from week one.

Method first

Baselines, splits and spreads before any model.

Student-owned

The question, code and words stay the student's.

Rules on the table

We read each program's AI rules together.

Across the US

Live video, nothing to drive to.

Two formats

A batch of fifteen to twenty, or one to one.

Student work

Work from our teenage students

Projects that started as practice and grew. The student labs page has more.

NutriLife AI nutrition coach project screenshot

AI and ML

NutriLife

An AI nutrition coach that reads what you eat and works you toward a target.

by Bhavya · Open it

Misti AI chatbot for maths and coding screenshot

AI and ML

Misti

A chatbot that answers mathematics and programming questions, built and deployed by a student.

by Harshit · Open it

GuardianX AI internet safety assistant screenshot

AI and ML

GuardianX

An assistant that helps a young person recognise unsafe situations online.

by Vivaan · Open it

SkyCast weather forecast application screenshot

Web app

SkyCast

A weather forecasting site with live conditions for any location.

by Krish

Fees

Fees

Programs and competitions set their own terms; we charge only for teaching. Payment is monthly in US dollars, with no long contract, and you can stop when a month ends.

Free first class

USD 0

no card required

  • A free first class
  • Bring your research idea
  • No card details
Book it

Group batch

USD 100

a month, billed in US dollars

  • Fifteen to twenty students
  • One regular teacher
  • Choose your batch
  • Monthly
Start here

One to one

USD 150

a month, billed in US dollars

  • A teacher to yourself
  • Paced to your deadline
  • Times at your convenience
Enquire

What families say

Rated 4.9 across 547 Google reviews

Parents' Google reviews, unedited.

★★★★★

"The one step solution for my son. Modern Age Coders make learning coding so simple that kids love it. The teachers explain complex concepts clearly with practical exercises and interactive content."

Ria Mukherjee

Parent

★★★★★

"Modern Age Coders has been a game-changer for me. I struggled to grasp IT concepts and coding before joining, but their classes transformed everything. I can now confidently write complex programs with ease."

Samriddha Mondal

Student

★★★★★

"One of the most wonderful education centres out there. Education is not limited to school syllabus but focuses on skill development."

Vansh Agarwal

Student

★★★★★

"My child Dhairya is really enjoying the Modern Age Coders classes. This is his first online class and he eagerly looks forward to it. I can already see his improvement, and the teachers are very cooperative."

Sonam Oswal

Parent of Dhairya

★★★★★

"Modern Age Coders have wonderful teachers who teach in a clear, easy and practical way. The teacher boosts students' confidence and inspires them to learn without hesitation."

Sonu Goyal

Parent

★★★★★

"I highly recommend this computer coding class! The teachers are incredibly knowledgeable and passionate about coding."

Ritu Kedia

Parent

Questions from families

AI research programs for teenagers: common questions

What are AI research programs for high school students?

They are programs where a student's own research is mentored or judged. In the US, Regeneron STS and Regeneron ISEF judge finished projects, while RSI and BU RISE place juniors in summer research. Machine learning work fits all of them.

When is the Regeneron STS 2027 deadline?

The 2027 rules say applications and recommendations are due on Thursday, November 5, 2026 at 8:00 pm Eastern Time.

Can I use ChatGPT or other generative AI in an STS project?

Only in limited ways. The 2027 rules forbid generative AI for writing application answers, drafting the Research Report or generating citations. AI tools may be used in the research itself if disclosed and cited, and AI-written code needs a citation and a prompt log.

Which ISEF category is a machine learning project?

Usually Robotics and Intelligent Machines, subcategory Machine Learning, which Society for Science defines as the construction and/or study of algorithms that can learn from data. An app built around a model may fit Software Design instead.

Can a team enter Regeneron STS?

No. The rules say applicants must complete individual research projects, and students may not split a team project and enter as individuals.

Who can apply to RSI 2027?

CEE says high school juniors are eligible and seniors are not. Applications for RSI 2027 are open through 11:59pm on December 11, 2026, and in 2027 CEE is adding a site at Stanford for 50 scholars.

Does my teenager need a university lab to do AI research?

No. Many machine learning questions can be tested on a laptop with public datasets. STS also permits research done alongside adult researchers in an institution, as long as the student's own role is clear.

Will your classes help my child win?

We cannot promise any result, and we never write a student's research. We teach the methods, coding and statistics so the student can design and defend their own work.

Are you connected with Society for Science or CEE?

No. We are an independent online school. Every rule on this page comes from the organizers' own published pages, which are linked so you can check them.

When are classes for US students?

Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays.

How much do your classes cost?

The first class is free. After that, teaching is paid monthly, and the fees section of this page gives the figure in US dollars. Programs and competitions set any costs of their own.

Elsewhere on this site

Related pages

More for US high schoolers serious about computer science.

Coding classes in the USA

All our pages for American families.

Summer research programs

RSI, PRIMES and SSP in depth.

AI projects for high schoolers

A ladder from first model to research.

Hackathons

Build fast, present clearly.

Training ML models

The craft behind a fair experiment.

AI and ML course for teens

Our full teen machine learning track.

AI certification for high school students

Four AI and Python certification exams for US high schoolers, compared from the issuers' own pages.

AI summer programs for high school students

AI summer programs for US high schoolers in 2027, and an eight-week AI summer you can run at home.

AI literacy for elementary students

AI literacy for K-5: what it means, what standards expect, and two activities.

Start here

Book a free first class

Tell us the question your teenager is curious about, even if it is rough. The first class works out what they need to learn to test it themselves.

Want to see the teaching first? Read how we teach or browse all courses, each with its syllabus.

WhatsApp us · +91 91233 66161 · connect@modernagecoders.com

Our teachers are based in India and teach on live video, which is how we reach students in every US state. WhatsApp gets the quickest answer.

Free first class; no card needed.

Want a class sooner? Book a priority demo

A priority demo is a full class of about 45 to 60 minutes. After the class the mentor sends a written skill report and a personal learning roadmap on WhatsApp. If you enrol, the demo fee is adjusted against your first month's fee. Share your demo Order ID with us on WhatsApp and we take it off.

Book a priority demo

Keep exploring Modern Age Coders

Related courses and guides

Learn more

Free resources

From the blog

Start here

Free classPriority demoCall us