The Project Index.

Vibha rewrote the ghost logic three times. The first version worked and she threw it away because she couldn't explain why it worked — which is the standard we ask for.

Advait built the level timer first, before any level existed, because he wanted to know if the idea was fun before committing to it. Prototyping instinct, at twelve.

Pradyun reverse-engineered a game he loves. Each attack phase is its own state, and he added the third one only after deciding the second was too easy — playtesting on himself.

Arham chose a problem from his own house. Modelling a car as an object came easily; deciding what belonged in the class and what didn't took considerably longer.

Monal writes. This started as a story and became a program, and the branching structure came out of the plot rather than being imposed on it.

Nishal plays. He wanted an assistant that tests you rather than tells you, and spent the last two sessions tightening prompts until the questions stopped being trivially easy.

Saanvika was writing a novel and kept stalling on names. The tool exists because the problem was real, which is why she finished it.

Prateek tested this on his own family before he'd let anyone else see it. The feedback loop he built — ask, listen, critique — mirrors the way he'd been taught.

Hansika trained the classifier on real messages and discovered it was confidently wrong about promotional texts. Watching a model fail is how you stop trusting one blindly.

Arham trained the model with his own voice, then found it didn't recognise his sister's. That discovery — about who a model is trained for — was the whole lesson.

Start where the curiosity already is. Aarna analyses Roblox because Aarna loves Roblox, and along the way she learned what a dataset is and why a chart needs labels.

Mohnish arrived with a question about whether surfaces favour certain players, and had to learn enough pandas to answer it honestly rather than convincingly.

Nishal's second dataset. This time he cleaned the data himself and found the missing values before we pointed them out.

Ayush chose the largest dataset he could find, then learned the hard way that a large dataset is mostly cleaning. He kept going, which is the part worth reporting.

The generator was the difficult half — Nishal had to work out how to produce a puzzle with exactly one solution, which is a genuinely hard problem.

Prithevij knows a great deal about Batman. The quiz questions were written before the page was, and the layout was built to serve them.

Anish rebuilt something from before he was born. Handling the tail growth cleanly took him from copying an approach to understanding one.

Ria's letter-colouring logic is subtler than it looks — duplicate letters break the naive version, and she found that case herself while testing.

Aarnav built this for a problem in his own neighbourhood. The design questions — who sees what, and what should never be shown — took as long as the code.

Hania brought her science homework to a coding class and asked whether she could turn it into a game. That instinct is exactly what we hope shows up by this age.

Varun built the quiz before the explanation, then realised the explanation had to teach whatever the quiz asked. Working backwards from the test taught him the structure.

Aarna's earliest project here. Everything she has built since is visible in embryo in this one — pick a subject you like, explain it, then test whether it landed.

Mohnish wrote this while preparing for AP Computer Science. The syllabus wanted a class exercise; he wanted a thing that worked on real rates, and did both.

Sumedh's first program that remembered anything after it closed. Reading and writing files is where a beginner's mental model of a program changes shape.