Help shape challenges by submitting needs
Challenges are built from needs submitted by the people who meet them. We are collecting a hundred or more, as different from one another as they can be.
🏆 There is at least CHF 500 in vouchers to win — travel to the hackathon in November, or AI and compute credits for a local winner.
Register to submit
Submitting takes an account: it is how we can write back to you, connect your needs to your application, and credit your referrals.
UNIGE, EPFL, HUG, Wyss Center, FCBG and CHUV addresses are admitted immediately.
What counts as a need
- One sentence. Concrete enough that someone could start on it in a weekend.
- A problem, not a solution. If you already know the answer, describe the gap it fills.
- Yours. A need you have actually run into is worth more than a plausible one.
Leaderboard
- needs submitted11
- contributors2
- distinct directions11 / 100
Most original so far Provisional
- thibaut.chataing7824
- wazaahhh4364
Points reward distance from everything already submitted. More needs earn more, with diminishing returns — the full formula is below.
These standings are indicative, not final. Points are scored against everything submitted so far, so a rank moves as the pool grows. Before the call closes, every submission is reviewed for validity — anything that is not a genuine need statement is removed and the board is recomputed. Ranks and totals shown here can change.
What we are asking for : 100 needs, as different as possible
The number is the point
A hackathon can be built from a dozen needs. A map of what is missing cannot. We are aiming for a hundred or more, from as many vantage points as we can reach — ward, clinic, lab, bench.
Quantity only counts if the needs differ
A hundred restatements of the same three problems is an echo, not a map. So the score rewards distance: every submission is measured against everything already submitted, and what earns most is ground nobody has covered.
We measure the spread, and it holds up
We developed a robust method for this, validated on an earlier library at this scale: every need becomes a sentence embedding, compared with every other. Mean pairwise distance reached 0.70 of 1 — broad coverage, few duplicates.
Enough needs, and the collection becomes an instrument
With enough spread the library stops being a list and starts answering: which needs recur across wards, and where digital phenotyping in mental health is still empty. A dozen needs cannot show an empty region; a hundred can.
It is written blind, on purpose
Nobody sees the pool while the call is open. That is what makes the originality score mean something: you are describing what you see, not differentiating yourself from what others already wrote.
The library outlives the hackathon
It stays as a standing record of unmet need for the Faculty — feeding grant proposals, the next hackathon, and the digital phenotyping platform at Campus Biotech long after November.
How points are calculated
The formula is public and identical for everyone. It pays for needs that differ from what has already been submitted, and it pays for submitting more of them — with diminishing returns, so no one can win on volume alone.
points = base × originality × order × AI-check × documentation
| Factor | What it measures | Range |
|---|---|---|
| Originality | How far your need sits from the nearest need already in the pool, as the semantic distance between the two sentences. Computed at the moment you submit, against a pool you cannot see, and never revised afterwards. | 0 → 1 |
| Order | Your first five needs each count in full. Every further need counts 0.7× the one before it, and nothing counts past a hundred. More needs always earn more; farming the board with near-copies earns almost nothing. | 1 → 0 |
| AI-check | A need a language model would produce from the theme alone counts half, or nothing at all if it is a near-copy of one. That baseline is generated before the call opens, so everyone is measured against the same machine. | 0, ½ or 1 |
| Documentation | The five brief fields — first idea, existing data, skills, constraints, evaluation. Statement only counts 1.0, a complete brief 1.5. The one factor you can still raise after submitting. | 1 → 1.5 |
- The base is a constant for the whole call, the same for every need: it sets the scale and never the ranking.
- Your standing is the sum over all your needs, which is where quantity and diversity multiply. Five needs pointing in five directions are worth far more than twenty variations on one — exploring new territory is simply cheaper than restating the territory you already covered.
- Points shown during the call are provisional. They are scored against everything submitted so far, so a rank moves as the pool grows, and every submission is checked for validity before the call closes — anything that is not a genuine need statement is removed and the standings are recomputed.
- When the call closes the organisers may apply a final expert factor to individual needs. It adjusts the ranking; it is applied by people reading needs, not by the model.
Originality is an indicator, not a verdict
The meter tells you where your need falls relative to the pool. It does not tell you whether the need is worth having, and a need that echoes others is emphatically still worth submitting. When several people who cannot see each other describe the same need, that convergence is evidence the need is real and widely felt — the engine records it as a cluster, and a large cluster is among the strongest signals the library produces. Stickiness is measured, not scored: it earns no points and it is one of the things we are most looking for. Write the need you actually meet, and let the meter say what it says.