United States · Grades 6 to 8 · Projects with real code
AI projects for middle school students
The difference between a middle schooler who has heard about AI and one who understands it is usually a single project: one they built, broke and fixed themselves. The right project for grades 6 to 8 is small enough to finish in a few weeks, needs no adult accounts, and teaches one real idea, such as learning from examples, or why a model trained on lopsided data gets things wrong. Below are ten such projects, two of them written out in full Python with the output we got, and the route by which a good one can become a science fair entry.
Live teaching since 2020 · 10,000+ students · Projects your child writes, not ones we write for them
The short answer
Strong AI projects for middle school students teach one real idea each: a rock-paper-scissors bot that learns a friend's habits, a message sorter that learns topics from examples, a Scratch game trained in Machine Learning for Kids, or AI for Oceans. Modern Age Coders runs live online project classes for grades 6 to 8 (ages 11 to 14) in Python and Scratch, in a batch or one to one.
Build with a teacher
Where these projects get built, with a teacher
Each is live, with a regular teacher who helps a student finish what they start.
Our list, ordered from a first afternoon to a science fair entry. The Scratch and browser projects come from free tools described on their own pages; the Python ones are written out in full further down.
AI projects for middle schoolers (our selection; tool descriptions quoted from each tool, read 17 October 2026)
Project
Tool
The idea it teaches
Grade
AI for Oceans
CodeAI, in the browser
Training: "train a real machine learning model to identify sea creatures and trash in the ocean"
6
Rock, Paper, Scissors
Machine Learning for Kids and Scratch
Images: "Make a Rock, Paper, Scissors game in Scratch that learns to recognise hand shapes."
6 to 7
Journey to school
Machine Learning for Kids
Prediction: "Train the computer to be able to predict how you travel to school in the morning."
6 to 7
Smart Classroom
Machine Learning for Kids and Scratch
Language: a smart assistant that controls virtual devices
6 to 8
Top Trumps
Machine Learning for Kids and Scratch
Numbers: "Train a computer to be able to play the Top Trumps card game in Scratch."
7
Fooled
Machine Learning for Kids
Bad data: how computers make mistakes if they are trained badly
7 to 8
Rock-paper-scissors bot
Python, no installs
Learning a habit from history (project 1 below)
7 to 8
Sport or music sorter
Python, no installs
Learning words from examples (project 2 below)
7 to 8
Headline testing
Machine Learning for Kids and Scratch
Testing: "Create a testing tool in Scratch to measure how good a machine learning model is."
8
Chatbot on a topic
Machine Learning for Kids and Scratch
Questions: a chatbot that answers questions about a topic you choose
8
How to choose one
Pick by the idea, not the tool. A sixth grader who has never trained anything learns most from AI for Oceans or Rock, Paper, Scissors, where the training step is visible and quick.
Seventh and eighth graders get more from the Python projects, because nothing is hidden. Every line that does the learning is on the screen, and they can change it.
What makes a project count
It is finished, it was tested on examples it never saw, and the student can say why it got something wrong. Those three together are what a teacher or a science fair judge looks for.
A project that only works on the examples it was trained on has not learned anything yet. Testing on new examples is the step to make sure is never skipped.
A rock-paper-scissors bot that learns its opponent
This bot keeps a record of every move a friend has made, finds the friend's favorite, and plays whatever beats it. It is the simplest honest example of learning from data. It is our own script, built only on the collections module that comes with Python, and its output below is from a run on 17 October 2026 with Python 3.13.
Project 1 · the learning botpython
# Project 1: a rock-paper-scissors bot that learns (ours)
from collections import Counter
BEATS = {"rock": "paper", "paper": "scissors",
"scissors": "rock"}
def bot_move(history):
if not history:
return "paper" # no data yet: just pick
favorite = Counter(history).most_common(1)[0][0]
return BEATS[favorite] # beat their usual move
friend = ["rock", "rock", "paper", "rock", "scissors",
"rock", "rock", "paper", "rock", "rock"]
wins = 0
for turn, move in enumerate(friend):
bot = bot_move(friend[:turn])
if BEATS[move] == bot:
wins += 1
print("bot won", wins, "of", len(friend))
Output on our run: bot won 7 of 10, which is 7 / 10 = 70%. Python describes a Counter as a dict subclass for counting hashable objects, and most_common(1) returns the single most frequent move with its count.
Why it wins
Our pretend friend plays rock seven times out of ten. After a couple of turns the bot has seen enough rocks to play paper almost every time, and paper beats rock.
That is the simplest kind of prediction: find the most common pattern in the past and bet on it continuing. When people are predictable, it works well.
How to beat it, and improve it
Play a different move every turn and the bot's favorite keeps changing; its wins drop. Ask the student to try, and to explain why.
Then improve the bot: let it look only at the last five moves, or at what the friend plays after a win. Each change is a small experiment with a number to compare.
Project 2, in Python
A message sorter that fails, and the one-line fix
This program learns which words belong to sport and which to music from six example sentences, then sorts four new ones. The first version gets half of them wrong, which is the most useful thing it does. Our script again, run the same day.
Project 2 · sport or music, before and afterpython
# Project 2: sort messages into sport or music (ours)
from collections import Counter
training = [
("the team scored a goal in the last minute", "sport"),
("our coach said the match starts at noon", "sport"),
("she ran the fastest race of the season", "sport"),
("the band played a new song tonight", "music"),
("he practiced guitar chords for the concert", "music"),
("that album has a great drum beat", "music"),
]
tests = [("the goal came late in the match", "sport"),
("a quiet song on the piano", "music"),
("the drummer joined the band", "music"),
("the race was on a rainy day", "sport")]
def train(skip=()):
words = {"sport": Counter(), "music": Counter()}
for text, label in training:
words[label].update(w for w in text.split()
if w not in skip)
return words
def classify(words, text):
score = {label: sum(words[label][w] for w in text.split())
for label in words}
return max(score, key=score.get)
for skip in [(), {"the", "a", "of", "in", "on", "at"}]:
words = train(skip)
right = sum(classify(words, t) == label
for t, label in tests)
print("skipping", len(skip), "words:", right, "of 4")
Output on our run: skipping 0 words: 2 of 4, then skipping 6 words: 4 of 4. The only change between the two runs is the set of words the program ignores.
What went wrong first
The sport sentences happen to use "the" more often. So in the first run, any new sentence full of "the", such as the drummer joined the band, scores higher for sport. The model learned the wrong pattern.
This is a real problem in real systems. Common words like the and a carry almost no meaning but appear everywhere, and they can drown out the words that matter.
What the fix teaches
Telling the program to skip six very common words lets the useful ones, band, song, goal, race, decide. The score goes from 2 of 4 to 4 of 4.
Four test sentences are far too few to prove much, and a curious student should say so. Writing ten more tests, including some tricky ones, is the natural next step.
Run both versions. Check you get 2 of 4, then 4 of 4.
Add a third topic. Try food, with three example sentences, and two tests.
Write tricky tests. "The band played at the game" belongs to which? Decide first, then run it.
Count the evidence. Print the score dictionary for one sentence to see why the model chose.
Write it up. One paragraph: what it does, where it fails, what you changed.
Without typed code
AI projects in Scratch with Machine Learning for Kids
For sixth graders, or anyone not ready for typed code, Machine Learning for Kids connects real training to Scratch. Its own description: "This free tool introduces machine learning by providing hands-on experiences for training machine learning systems and building things with them."
How Machine Learning for Kids works, from its published English text (read 17 October 2026)
Point
What the site says
Step 1
"Collect examples of things you want to be able to recognise"
Step 2
Use the examples to train a computer to recognize them
Step 3
Make a game in Scratch that uses what the computer learned
Without an account
"Try without registering"; with Try it now you can only create one project at a time
How long it has run
The tool was first made available in 2017
Project worksheets
Each project has a downloadable worksheet; we counted 63 titled worksheet entries in its English text
A good first Scratch project
Rock, Paper, Scissors suits a first try: the student photographs their own hand shapes as examples, trains, and then plays. When the game misreads a hand, the fix is obvious and physical, more examples, better lighting.
The Fooled worksheet then makes the point deliberately. Its description is "Learn about how computers can be confused and can make mistakes if they're trained badly", and it is the lesson most worth having by grade 7.
Moving on to Python
Scratch projects hide the learning step inside the tool. Once a student asks how it works inside, they are ready for projects 1 and 2 above, where the counting is on screen.
A good order is one project of each kind: a Scratch one to see AI working, then a Python one to see how.
Science fair
Turning an AI project into a science fair entry
The national competition for grades 6 to 8 is reached through science fairs affiliated with Society for Science; ask whether your local fair takes computer science entries. Here is the route as Society for Science describes it, then our template for the write-up.
The Thermo Fisher Scientific Junior Innovators Challenge, from Society for Science (read 17 October 2026)
Point
What Society for Science says
Who
Students in 6th, 7th and 8th grade in the United States
How to enter
Students must first compete at a Society-affiliated science fair
Nomination
The top 10% of the 6th, 7th, and 8th grade competitors at Society-affiliated fairs in the U.S. are eligible for nomination
2026 numbers
1,997 entrants, a Top 300, and 30 finalists
2026 Finals Week
October 23 to 28, in Washington, DC
Making an AI project fair-ready
A fair project needs a question, not just a program. "Can a bot beat my sister at rock-paper-scissors?" becomes "Does a bot that remembers only the last five moves beat one that remembers all of them?"
Then it needs a fair test: the same opponent moves for both bots, enough games to mean something, and the results in a table. Project 1 above is most of that already.
The write-up, in five parts (our template)
Question. Prediction before testing. Method, including the code and the test examples. Results, with numbers. What went wrong and what you would change.
The fifth part matters most. A judge learns more from a student who can explain why the sorter failed than from one who only shows it working.
Our rough check of the published 2026 numbers: 1,997 x 0.15 is about 300, so the Top 300 was roughly the top 15% of entrants. That is our arithmetic on Society for Science's figures. Your school or district fair is the first step; ask a science teacher which affiliated fair it feeds. Source: Society for Science, Thermo Fisher JIC. We are not connected with Society for Science or Thermo Fisher Scientific.
For parents and teachers
Helping without doing it for them
Middle schoolers need help to finish projects, and the help is where projects go wrong. These are the rules we teach by.
Ask, do not type
When a student is stuck, ask what they expected and what happened. The fix they find themselves is the one they remember.
Keep it small
A finished project with four test examples beats an ambitious one that never runs. Scope down early and often.
No adult accounts
The Python projects here need nothing but Python. For any online tool, read its terms for the child's age first.
A weather forecasting site with live conditions for any location.
by Krish
Fees
Fees
Every project on this page can be built for free at home, and we say so. Our classes add a teacher, billed monthly in US dollars with no long contract; stop after any month.
"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 and teachers
AI projects for middle schoolers: common questions
What is a good AI project for a middle school student?
One that teaches a single real idea and can be finished in a few weeks: a rock-paper-scissors bot that learns habits, a message sorter that learns topics, a Scratch game trained in Machine Learning for Kids, or AI for Oceans.
Can a sixth grader build an AI project?
Yes. Browser and Scratch tools let a sixth grader train a model with no typed code. CodeAI's AI for Oceans has students train a real machine learning model to identify sea creatures and trash in the ocean.
Does my child need an account for these projects?
Not for the Python projects on this page, which need only Python. Machine Learning for Kids offers a Try without registering option, limited to one project at a time.
Can an AI project go to a science fair?
Yes, if your school or district fair accepts computer science entries; ask your science teacher. For grades 6 to 8, the Thermo Fisher Scientific Junior Innovators Challenge is reached by competing first at a Society-affiliated science fair.
How do students qualify for the Junior Innovators Challenge?
Society for Science says the top 10% of 6th, 7th and 8th grade competitors at its affiliated fairs in the US are eligible for nomination. In 2026 there were 1,997 entrants and 30 finalists.
Why did the message sorter get answers wrong at first?
Because common words like the appeared more in the sport examples, so any sentence full of them looked like sport. Skipping six common words fixed it, which is a real technique in language AI.
Should I help my child with their AI project?
Help by asking questions, not by writing code. Ask what they expected and what happened. A project the student can explain is worth more than a polished one they cannot.
Do you build projects for students?
No. We teach the ideas and the code so students build their own projects. That is also what science fairs expect.
When are classes for US middle schoolers?
Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays.
What do your project classes cost?
The first class is free. After that, families pay monthly, and the fees section of this page gives the figure in US dollars.
Our teachers teach from India over live video, so any US family can join from home. WhatsApp is the quickest way to reach us.
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.
