Skip to content
    2Morrow Lab

    Build, make, research

    What others have built with AI

    Four shelves of real projects. Each one says what the AI did and where it falls short. At the end of each shelf is a story where something went wrong.

    People your age

    Students who built something real.

    Karl Hechema, 16, Dubai

    A free revision site for the French Brevet exam.

    What the AI did
    It helped write the code and makes the quizzes.
    Where it falls short
    The user numbers are his own. The hard part came after launch.

    Source: Gulf News · October 2025

    Arush Shangari, 17, Massachusetts, USA

    A program meant to warn of floods on one river 24 to 72 hours ahead.

    What the AI did
    Models he trained himself read public satellite images to do this.
    Where it falls short
    The 98% accuracy is given without a method. It knows only one river and is not in use yet.

    Source: Boston.com · August 2026

    Liam Blignault, Grade 11, Gqeberha, South Africa

    A study app for the computer that works without the internet.

    What the AI did
    A small language model runs offline on a computer with 4 GB of memory and answers in a chat.
    Where it falls short
    The source does not say which model it is or how good its answers are. It is a school project and not public.

    Source: Nova News · March 2026

    Close to home: Germany and Europe

    School students, apprentices and university students in Europe.

    Magnus Schlinsog, Jugend forscht 2026 (German science fair)

    Jacob, a chatbot that answers in Easy German (Leichte Sprache).

    What the AI did
    It writes the answers. He retrained it for this on about 1,000 sample conversations.
    Where it falls short
    His rule filter still misses errors. Answers can be slow.

    Source: heise online · May 2026

    Matching project: A college letter in plain language

    Apprentices from the Lindau region in Bavaria, at an AI challenge

    An image recognition model that finds screws, nuts and washers.

    What the AI did
    A pretrained model learned from images they collected and labelled themselves.
    Where it falls short
    It was a guided workshop with a set task. The model is a practice model for class.

    Source: IHK Schwaben · July 2026

    Girls and non-binary young people aged about 15, TU Darmstadt

    Five clickable app prototypes on social problems, built in three days.

    What the AI did
    An AI chat helped them research who the apps are for. They had to check every answer.
    Where it falls short
    The apps are prototypes and not released. The source says nothing about tests with users.

    Source: TU Darmstadt · August 2026

    Team Alpha, university students from Potsdam, Essex and Vietnam

    Adaptly, a website that reworks a text for people with dyslexia, ADHD, autism or deafness.

    What the AI did
    An AI restructures the text, adds questions to check understanding and shows sign language as animated stick figures.
    Where it falls short
    It has not been tested with the people it is meant for. The source does not name the AI it uses.

    Source: University of Potsdam · December 2025

    Matching project: A college letter in plain language

    Research and check

    Questions asked of real data, from stars to old letters.

    Volunteers in the Hubble Asteroid Hunter project

    They found 1,701 asteroid trails in old Hubble images.

    What the AI did
    Trained on the volunteers' labels, it searched the whole Hubble archive for more trails.
    Where it falls short
    A trail is not yet a confirmed new asteroid.

    Source: ESA/Hubble · April 2024

    Michelle Kukan, history student at Ramapo College, USA

    A test of whether AI can transcribe the old papers of Jane Addams.

    What the AI did
    An AI transcribed typed and handwritten pages. It was retrained with her corrections.
    Where it falls short
    Her verdict is that it is not good enough yet, above all for handwriting. She measured the times herself.

    Source: Ramapo College · 2025

    A computer science class at a high school in California

    A fact-checker that listens to a live debate and checks the claims.

    What the AI did
    One AI turns the speech into text, and a second one checks the claims.
    Where it falls short
    The 87% hit rate comes from one single debate and was measured against the class's own checks.

    Source: GovTech · November 2024

    Luke Farritor, Youssef Nader and Julian Schilliger, students

    They made text in unopened scrolls from Herculaneum readable and won the Vesuvius Challenge grand prize.

    What the AI did
    An AI found ink in CT scans of the scrolls.
    Where it falls short
    The team built on years of work by others, on scanning and on reading old texts.

    Source: Vesuvius Challenge · 2024

    No coding: build and make

    A teacher, a musician, a screen-reader user and a choir-school team make things with AI.

    David Webb, teacher at an international school

    An app that suggests books from the school library to his students.

    What the AI did
    It wrote the code from his description.
    Where it falls short
    Once, files were lost with no backup. He says it works best alongside teachers.

    Source: EdSurge · In their own words · June 2026

    A team from the Regensburger Domspatzen choir school, German AI competition 2025

    A study of Gregorian chants with AI, and new pieces in their style.

    What the AI did
    It helped study the chants and make new pieces in their style.
    Where it falls short
    The source says little about how the team worked. It does not say whether the pieces were performed.

    Source: Tübingen AI Center · November 2025

    Priscilla Angelique-Page, musician and sound engineer in Nottingham, UK

    Two albums by an invented band, the Intelligent Band Machine.

    What the AI did
    AI tools made the music, the voices and the band's pictures. She wrote all the lyrics herself.
    Where it falls short
    For an album of 14 songs, she generated about 1,000. Who owns the music is still an open question.

    Source: Singularity Hub · In their own words · May 2025

    Aaron Preece, screen-reader user with almost no coding experience

    A small program that reads out subtitles next to a streaming video.

    What the AI did
    One AI chat wrote the plan, and a second one wrote the code.
    Where it falls short
    Syncing it with the video automatically did not work. A simpler version controlled by keys did.

    Source: AccessWorld · In their own words · 2026

    Matching project: A college letter in plain language

    Are you in one of these stories and want it removed? Write to us and we will take it down. info@2morrowlab.org

    Try it first, then decide.

    Finding your next step is a skill you can learn. How to make yourself a new plan isn't in any curriculum. That's what we practise with you.