Before you start · read this first
Selecting a project
There is not one right way to be an innovator. Some students code, some run experiments, some build a physical product and some write. Below are the main paths students take on Immanuel Innovation, what each one actually involves week to week and how to use AI as a genuine thinking partner along the way.
Whatever you pick, the goal is the same: ask a real question, do real work to answer it and end up with something you are proud to show someone.
Every category below shares the same shape: notice something worth investigating, do the work properly, then show what you found or built. What changes is the type of evidence you produce - working code, a tested prototype, a written argument or a result from an experiment.
Read through the options, then pick the one that fits the itch you already have. If you are not sure yet, that is normal - browse the project library once you have registered and see what grabs you.
The main types of project
Ten broad paths. Most of the 100+ briefs in the library are a specific version of one of these.
Build an app or digital tool
Start from a problem you or people around you actually have - not a feature you think would be cool. Sketch the screens on paper first. Build the smallest version that works, get real people to try it, then fix whatever confused them.
AI helps by turning your sketch into working code, explaining error messages line by line and stress-testing your app in ways you would not think to try.
Scientific research
Ask a question that can genuinely be tested, gather evidence and see whether it holds up. There are two distinct routes into this - a hypothesis-led approach and a data-driven approach - both walked through in full in the deep dive below.
AI helps by suggesting testable hypotheses, finding related studies fast and checking your statistics once you have results.
Invent a product
Spot a genuine gap or annoyance. Sketch several rough solutions before committing to one. Build a cheap, rough prototype - cardboard, code or clay - and put it in front of real users as early as possible.
AI helps by checking existing patents so you are not reinventing something, suggesting materials or mechanisms and drafting the technical write-up once it works.
Article or white paper
Pick a question or a claim worth digging into. Go beyond the first page of search results - find primary sources, and interview someone who actually knows the topic if you can. Structure an argument built on evidence.
AI helps by widening your reading and summarising sources quickly - but the argument and the sentences should end up genuinely yours.
Business or social enterprise
Find a real customer with a real problem they would pay to solve. Work out a simple model - what you sell, to whom, for how much. Test the idea on a handful of actual people before spending anything on building it.
AI helps by running quick market research, drafting a business plan and modelling numbers - "what happens if I sell 50 a month?"
Game design
Decide what makes the game fun before you decide what it looks like - that is the core loop. Build the simplest playable version, even on paper, and playtest it constantly, changing whatever rule is not working.
AI helps by generating dialogue and lore ideas, writing game logic in code and standing in as a playtester when you are stuck alone.
Board game or TTRPG
Same instinct as game design, played out on the table. Draft rules on index cards, prototype with objects you already own and playtest with friends or family, tightening the rules after every round.
AI helps by checking whether one strategy always wins, generating flavour text and formatting a printable rulebook.
Screenplay, novel or graphic novel
Build characters who want something, and something in the way of it. Outline the shape of the story before writing scene by scene. Get a full rough draft down before you polish a single sentence.
AI helps by talking through plot problems and pacing - but keep the voice yours, since readers can always tell.
Documentary, film or podcast
Choose a story or topic where real people or footage are actually available to you. Write a script outline or storyboard, then film or record in short segments you can edit together afterwards.
AI helps by transcribing interviews, drafting a script outline and suggesting a shot list before you film.
Campaign or advocacy project
Pick a cause you genuinely care about. Set one clear, achievable goal - "get 200 signatures" beats "save the world." Then plan exactly how you will reach the people who can act on it.
AI helps by researching the facts behind your cause and drafting persuasive messaging you can then make your own.
Deep dive: doing real scientific research
This is the category students find hardest to start, so here is exactly how to do it. There are two genuinely different ways in, and good projects usually end up using a bit of both.
Hypothesis-led research
- Start with a hypothesis - an educated guess, grounded in existing science or plain common sense, that you can actually test.
- Try to disprove it, not prove it. That is the real heart of the scientific method: a hypothesis only earns your trust once it has survived a genuine attempt to knock it down.
- Design a fair test - decide what you will measure, who or what you will compare and what would count as evidence either way, before you collect anything.
- Collect your data, then run the numbers to see whether the difference you found is a real pattern or just chance.
Example hypothesis: "Home-schooled children have a higher reading age than children in day school." You would test reading age across both groups, control for things like age and compare the results.
Data-driven research
- Instead of starting with a guess, start with a large pile of data - one you find already published, or one you collect yourself through a survey or measurements.
- This route usually needs a lot of data before any real pattern is visible above the noise, so plan for volume from the start.
- Explore visually before you conclude anything - plot the data in different ways and see what shapes emerge.
- Pressure-test any pattern you spot: could it be explained by something else entirely, rather than the story you want it to tell?
Example: gather exam results, sleep habits and screen time across a year group, then look for which factors actually move together - without assuming the answer in advance.
What is a p-value, roughly?
Once you have collected results, statistics tell you whether a difference is likely to be real or just random noise. A p-value is a number that captures this - small values (conventionally under 0.05) suggest the pattern is unlikely to be down to chance alone, so your hypothesis has survived the attempt to disprove it. It does not prove you are right; it just tells you the result is worth taking seriously.
You do not need to be a statistician to do this properly - free tools (and AI) can run the test for you once you understand what question it is answering.
Use it to brainstorm testable hypotheses from a topic you care about, to find related studies so you are not starting blind, to spot weaknesses in how you have designed a test before you run it and to write or explain the code that runs your statistics. Let it accelerate your thinking - but the question, the data and the conclusion should be genuinely yours.
Most strong projects do not sit neatly in one lane. It is completely normal to start with a hunch, gather data to test it and end up spotting an unexpected pattern you were not even looking for along the way.
Ready to choose yours?
Browse the full project library, or start from a blank page and shape your own brief from one of the paths above.