Hackathons
Kaggle-style challenges run end to end inside sciFi notebooks — a shared dataset, a live leaderboard, and every submission a real, runnable notebook rather than a screenshot of one.
Our next challenge moves to financial data — the problems where severe class imbalance, hard latency budgets and explainability requirements all bite at once, and where a good AUC is emphatically not the same thing as a good model.
Catch fraud in a heavily imbalanced transaction stream, scored on precision at a fixed alert budget rather than raw AUC.
Build a scorecard that is both accurate and defensible — monotonic constraints and reason codes count toward your score.
Extract structure from messy financial text under a strict compute budget.
Forecasting and anomaly detection on transaction and balance histories.
Tracks and prizes are still being finalised. Tell us you want in and we'll email you the moment registration opens.
Notify meOpens your mail app — or write to lucky@scifi.ink
Builders competed to detect anomalies in ICU patient vital signs — a noisy, heavily imbalanced time-series problem where the cost of a missed anomaly and the cost of a false alarm are wildly asymmetric.
We host private hackathons on sciFi for companies and universities — your dataset, your problem statement, your leaderboard. Because everything runs in the browser, there is no environment setup to support on the day, which is usually the single biggest cause of a bad hackathon.
Preparing for the fintech challenge? Our blog covers fraud detection, credit risk and small language models in depth.