Germany · Bundeswettbewerb Künstliche Intelligenz · Sekundarstufe I and II
Bundeswettbewerb KI (BWKI) project help
The Bundeswettbewerb Künstliche Intelligenz asks pupils to pick a real problem and attack it with machine learning, then explain it in writing, in code and in a short video. It is judged as a piece of research, and its rules are strict about one thing above all: the work must be the pupil's own, and every bit of help must be declared. This page explains the contest from its organisers' own rules and shows exactly where a mentor fits, and where a mentor must stop.
Live online since 2020 · 10,000+ students worldwide · your project, your code, our teaching
In short
BWKI project help at Modern Age Coders means live online mentoring in English for pupils in Germany who want to enter the Bundeswettbewerb Künstliche Intelligenz. We teach the Python, machine learning and honest evaluation a project needs, and review the pupil's own plan, code and write-up; the pupil does the project and submits it. A regular mentor follows each pupil's pace and accent, alone or in a group of fifteen to twenty.
Where BWKI pupils start
Three courses behind a BWKI project
Machine learning from first principles, the Python underneath it, and the German contests course with its AI project track.
What is the Bundeswettbewerb KI, and who can enter?
A national project competition in artificial intelligence for school pupils, organised by the Tübingen AI Center and carrying a recommendation from the Kultusministerkonferenz.
What it asks for
The organisers' conditions describe a self-chosen project that tackles a problem of social or scientific relevance with methods of machine learning. No topics are set. A project may train a model on a dataset, made by the pupil, taken from the internet or provided by a partner, or may build in generative AI such as language or image models.
Work can happen at home or at school. The BWKI states that it does not provide rooms or materials, and that participants are responsible for their own computer and internet connection.
Who may take part
Anyone in Sekundarstufe I or II may enter, solo or with up to three teammates. Entrants must have German citizenship, or live in Germany, or attend a recognised German school abroad. Students may still enter in their first year of a first degree if they left school no more than a year before, and the same applies to young people doing voluntary service.
There are no age categories, the conditions say, but there are special awards for young talent.
Two lines in the conditions matter to families who speak English at home. The organisers accept a project written in English as readily as one in German. And residence in Germany is enough: a pupil at an international school who lives in Germany meets the residence rule on the organisers' own wording. The contest belongs to the AG bundesweiter Jugendwettbewerbe, the association of national youth contests, and the Minister-President of Baden-Württemberg is its patron.
The yearly rhythm
When does the BWKI run, and what were the 2026 dates?
Every year follows the same shape: register in spring, submit the idea in early summer, submit the project at the end of September, present in November.
The BWKI year, with the 2026 dates as the organisers published them
Stage
Usual timing
2026 date
Registration and team-building open
From early March
1 March 2026
Project idea submitted
By the end of June
12 July 2026, after an extension
Full project submitted
By the end of September
20 September 2026
Final, up to ten teams
November
13 November 2026, Tübingen
Sources: Bundeswettbewerb Künstliche Intelligenz, home page (bw-ki.de) and Wettbewerb page, read on 2 and 3 November 2026. The organisers publish each year's exact dates on the registration page; the 2027 dates had not been published on the pages read.
On the date of writing the 2026 round is almost over: the finalists present in Tübingen on 13 November 2026, and the organisers opened a public vote for the audience prize from 3 November. For a pupil reading this now, the realistic target is the next round, with registration expected from early March 2027. That leaves the winter to learn the methods, which is exactly when a mentor is most useful and least tempted to do the work.
Submission and judging
What must a BWKI project contain, and how is it judged?
Three parts, assessed by an interdisciplinary jury against seven criteria. Every team gets written feedback, finalist or not.
The three parts
Documentation: the idea, the methods, how it was built and what came out, plus possible sources of error and a critical view of the results.
Code: runnable and clearly structured, with every dataset, model, framework and outside tool marked.
Video: a pitch of two to four minutes that a non-expert can follow and an expert can still respect.
After submission
A selection jury reviews every project and invites up to ten teams to the final in November, where they present at a stand they design, before the public and the jury. The organisers cover finalists' travel and accommodation, and personal attendance is required for the prizes.
Teams not chosen still receive expert feedback, which the organisers suggest using to come back stronger the next year.
The seven BWKI judging criteria and the question behind each (English wording ours)
Criterion
The jury asks
Eigenständigkeit (own work)
How large is your own share of the project?
Originalität
How creative and new are the idea and the approach?
Technische Umsetzung
How demanding and reliable is the implementation?
Wissenschaftliches Arbeiten
Have methods, results and sources of error been examined traceably?
Praktische Relevanz
What use or insight does the project offer?
Codequalität
Is the code understandable, structured and traceable?
Präsentation
Are idea, method and results clear in the documentation and the video?
Source: Bundeswettbewerb Künstliche Intelligenz, Wettbewerb page (bw-ki.de/wettbewerb), read on 3 November 2026. The organisers add that the jury takes into account the prior knowledge, opportunities and resources each team had.
Allowed help
What help is allowed for a BWKI project, and how must it be declared?
Help is allowed. Hidden help is not. The conditions require a clearly visible, documented share of the pupil's own work, and full disclosure of every kind of support.
What the rules say
The organisers list what must be declared: the use of AI models and frameworks, the involvement of people, institutions or companies, datasets that were provided or downloaded, and any GPU computing power or technical infrastructure made available. On registering, participants also confirm that they solve the tasks and build the project themselves and will name every source and helper in the documentation.
The organisers say they will not use participants' ideas or results for their own purposes without asking.
Where we fit, and where we stop
Modern Age Coders has no tie to the BWKI or to the Tübingen AI Center. We teach machine learning, review a pupil's plan, explain why a piece of the pupil's own code behaves as it does, and ask the questions a jury will ask.
We do not choose the topic, write project code, write or edit the documentation, record the video or submit anything. If a pupil learned from us, that is help, and the pupil names us in the documentation, as the rules require.
The pupil proposes. The pupil brings a question and a possible dataset; the mentor asks what success would look like and how it could be measured.
The mentor teaches. Methods are taught on separate practice data, never by building the project for the pupil.
The pupil builds. All project code is typed by the pupil, in the pupil's own repository.
The mentor questions. Reviews are questions, not rewrites: why this split, what is the baseline, where could this be wrong?
The pupil declares. The final documentation lists every dataset, library, AI tool and person who helped, our classes included.
On AI tools the BWKI allows generative models inside a project, but every provider sets its own minimum age: under OpenAI's terms a teenager needs a parent's consent until 18, and Anthropic reserves Claude for adults, as recorded in our AI tools age guide. For pupils under 18 our teaching uses Python and open libraries, and we never put an adult-only tool in front of them.
A lesson of our own
Why is accuracy not enough? A lesson on evaluating a model honestly
The jury asks for analysis of error sources and a self-critical view of results. The most common trap in pupil projects is a single impressive accuracy figure. This example is ours, on invented data, to show why.
The made-up problem (ours)
Imagine 365 days of readings from a school weather station. On a few days there is heavy rain, and we want to warn the caretaker in advance using the morning humidity. The data below is invented by a seeded random generator for teaching. It is not real and not from any project.
The first 240 days are for learning; the last 125 are locked away for testing.
Two models
Always no: predict "no heavy rain" every day. It needs no data at all.
Threshold: predict heavy rain when humidity is at or above a number learned from the 240 training days, the one that most fairly balances catching rainy days against false alarms.
Two models and a confusion matrixPython
import random
random.seed(51)
days = []
for d in range(365):
heavy = random.random() < 0.07
humidity = random.gauss(88, 4) if heavy else random.gauss(71, 8)
days.append((round(humidity, 1), heavy))
train, test = days[:240], days[240:]
def scores(predict, data):
tp = sum(1 for h, y in data if predict(h) and y)
fp = sum(1 for h, y in data if predict(h) and not y)
fn = sum(1 for h, y in data if not predict(h) and y)
tn = len(data) - tp - fp - fn
accuracy = (tp + tn) / len(data)
recall = tp / (tp + fn) if tp + fn else 0
precision = tp / (tp + fp) if tp + fp else 0
return tp, fp, fn, tn, round(accuracy, 3), round(precision, 3), round(recall, 3)
def f1(t):
tp, fp, fn, tn, a, p, r = scores(lambda h: h >= t, train)
return 2 * p * r / (p + r) if p + r else 0
print("Heavy-rain days in test:", sum(1 for h, y in test if y), "of", len(test))
print("Always no: tp fp fn tn, accuracy, precision, recall =", scores(lambda h: False, test))
best_t = max(range(70, 96), key=f1)
print("Threshold learned on training days:", best_t)
print("Threshold model:", scores(lambda h: h >= best_t, test))
It prints Heavy-rain days in test: 10 of 125, then Always no: tp fp fn tn, accuracy, precision, recall = (0, 0, 10, 115, 0.92, 0, 0.0), Threshold learned on training days: 83 and Threshold model: (9, 8, 1, 107, 0.928, 0.529, 0.9).
The two models on the 125 test days of invented data (our calculation)
Model
Rain warned and came
False alarms
Rain missed
Accuracy
Rain days caught
Always no
0
0
10
92.0%
0 of 10
Threshold at 83
9
8
1
92.8%
9 of 10
Data: invented with random.seed(51) for this lesson, 365 days, heavy rain on about 7 percent. We chose the split, the threshold search and the seed; every figure was produced by that script under Python 3 on 3 November 2026.
Read the accuracy column alone and the two models look almost equal, 92.0 against 92.8 percent. Read the rest and they are nothing alike. The lazy model is right on 92 percent of days only because 92 percent of days are dry, and it misses every heavy-rain day. The threshold model catches 9 of the 10, at the price of 8 false alarms: its precision is about 0.53, so roughly every second warning is real. Whether that trade is good depends on the caretaker, and saying so in the documentation is exactly the scientific, self-critical work the BWKI criteria name. A pupil who reports only "92.8 percent accuracy" has hidden the most interesting part of their own project.
Run it. Write down the four numbers of each confusion matrix before reading the explanation.
Move the line. Fix the threshold at 80, then 86. What happens to false alarms and missed rain? Which would you choose, and for whom?
Break the rule. Learn the threshold on the test days instead. Why can that result not be trusted in a BWKI documentation?
Your project. Write one paragraph on which mistake would be worse in your own project idea, a false alarm or a miss, and why.
A plan
How could a pupil plan a BWKI project for 2027?
Here is our suggested plan, built backwards from the usual timing. The exact 2027 dates will come from the organisers.
A suggested plan for a 2027 BWKI entry (ours)
When
The pupil
In class
November 2026 to February 2027
Learns Python and data handling; keeps a list of questions they care about
Fundamentals, on our practice datasets
March 2027
Forms a team, registers, chooses a question with a dataset whose licence allows use
How to judge whether a question is answerable
April to June 2027
Builds a baseline, then a first model; submits the idea by the organisers' deadline
Evaluation, as in the lesson above
July to September 2027
Improves, documents error sources, records the video, submits
Questions on the draft, never edits to it
November 2027
If chosen, presents at the final; otherwise reads the jury's feedback
Practice answering hard questions aloud
The usual timing is from the organisers' Wettbewerb page; the plan and its contents are our own suggestion, read on 3 November 2026.
Class times
When can BWKI mentoring run in German time?
Late afternoons and weekends suit most teenagers. Our mentors teach from India, ahead of German time all year.
Sample mentoring times and the hour in India
Session starts (Germany)
Mentor's clock until 27 March 2027
Mentor's clock from 28 March 2027
16:50
21:20
20:20
18:15
22:45
21:45
Saturday 12:45
17:15
16:15
Sunday 16:00
20:30
19:30
Germany switches between UTC+1 and UTC+2 under the Einheitenzeitgesetz; winter time began on 25 October 2026 and summer time begins on 28 March 2027. India does not change its clocks (UTC+5:30), which puts the mentor four and a half hours ahead this winter and three and a half ahead next summer.
The table lists examples, not fixed batch times. One-to-one mentoring is usually weekly, with an extra session agreed as the submission date nears.
Milestones
From a first dataset to a project you can defend
Milestones we use with BWKI pupils to place them and to show parents where things stand.
BWKI project milestones (ours)
Stage
Roughly when
By then the pupil can
1. Data in hand
Month 1
Load a dataset, describe it, and state its source and licence
2. A baseline
Month 2
Name the simplest prediction and measure it
3. An honest test
Months 2 to 3
Split the data, train a model and read its confusion matrix
4. Error sources
Months 3 to 5
Find where the model fails and explain why
5. A defended project
Months 5 and up
Document, pitch and answer questions on every choice made
For pupils
If one number makes your model look perfect, look for the number that does not.
Write down every source and helper as you go. The rules ask for it at the end.
For parents
Ask to see the project folder, not the result. Your teenager's own commits are the clearest sign of real work.
Help is fine and allowed. Make sure it is all named in the documentation.
More courses
Nine more courses for AI-minded teenagers in Germany
All nine are taught live, in English, by one mentor from start to finish; book a seat in a group or a private slot.
A weather forecasting site with live conditions for any location.
by Krish
Fees
Fees for BWKI mentoring and AI classes
Eight live sessions a month make up the monthly fee. Our euro figure is converted with a reference rate published by the European Central Bank; what we actually charge is US dollars, and your bank converts them, which can leave a small gap between our euro figure and yours.
Google reviews from students and parents, quoted without changes.
★★★★★
"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
Pupils and parents ask
BWKI project help: common questions
Are the classes in German?
No. Mentoring happens live and only in English, and the mentor adjusts to each pupil's pace and accent. The BWKI accepts projects in German or English, so a pupil can also write the project in English.
Will you do the project or submit it for my child?
No. The BWKI requires a clearly documented share of the pupil's own work and the disclosure of every kind of help. We teach and ask questions; the pupil chooses the topic, writes the code and the documentation, records the video and submits.
Do we have to mention you in the project?
Yes. The BWKI conditions require all support by people, institutions or companies to be disclosed in the documentation. A pupil who learned from us names our classes there, as with any other help.
Who can enter the Bundeswettbewerb KI?
Pupils in Sekundarstufe I and II, alone or in a team of up to four, who have German citizenship, live in Germany or go to a recognised German school outside Germany. Some first-year students and people in voluntary service can enter if they left school within the past year.
When is the next BWKI round?
The organisers describe a yearly rhythm: registration from early March, the idea by the end of June, the project by the end of September and the final in November. In 2026 the final is on 13 November 2026 in Tübingen. Exact 2027 dates come from the BWKI.
May a teenager build a chatbot into a BWKI project?
The BWKI allows generative AI inside a project, provided it is declared. Providers set the ages, though: OpenAI asks for parental consent before 18, and Claude is for adults only under Anthropic's terms. We never use Claude with pupils under 18.
Are you connected with the BWKI?
No. We have no connection with the Bundeswettbewerb Künstliche Intelligenz or the Tübingen AI Center. We are an independent teaching company.
When are classes in German time?
Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays. Most pupils pick a late afternoon or a weekend.
How are fees paid?
Our euro fee is derived from a European Central Bank reference rate; the payment itself is taken in US dollars and converted by your own bank. The trial class costs nothing, and you enter no card.
Elsewhere on this site
Related pages
Further pages for young people in Germany who want to build with AI.
Bring a question you care about. In the first class the mentor asks how it could be measured, checks what you already know, and suggests what to learn before March. All of it happens in English, at a speed that suits you.
Mentoring is live online and only in English; our mentors are based in India. WhatsApp us in English for a quick reply.
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.
