VELUM

Velum

Catching water pollution before the evidence disappears.

The Problem Definition

The problem isn't a lack of expertise or willpower from environmental authorities. It's that the current process is structurally too slow to act within the window where proof still exists.

Right now, most pollution incidents go completely unattributed. This happens not because we can't find the source, but because we always arrive too late. By the time a report is filed and an investigation begins, the evidence has literally washed away.

The Solution

The satellite data was always free. The evidence window was always there. The only thing missing was a system fast enough to use both of them together.

Enter Velum.

Instead of waiting for human reports, Velum proactively monitors water bodies and alerts authorities the moment an anomaly appears. Users open Velum and see exactly what was found, what type of pollution it is, and the likely sources of the pollution. The evidence window remains open, giving authorities the critical time they need to act.

Our Solution / Method

Velum operates on a fast, three-step automated pipeline:

1. Detect

Sentinel-2 satellites don't just take photographs: they measure specific wavelengths of light that reveal chemical composition. One of those wavelengths is a direct proxy for chlorophyll concentration in water.

• Velum downloads satellite imagery the moment it becomes available.

• It computes the chlorophyll and turbidity readings for every pixel in the river basin.

• Any location where readings exceed the local baseline is instantly flagged. No human needs to notice anything first.

2. Classify

An algal bloom looks very different from an industrial discharge under satellite sensors, and the pollution type shapes everything downstream (which suspects are plausible, what evidence to look for, and how urgently to act).

• We use a visual AI model to examine a close-up of the affected area.

• It classifies the anomaly (e.g., algal bloom, industrial discharge, agricultural runoff).

• It outputs a pollution type, a severity rating, and a confidence score.

3. Attribute

Every registered facility upstream of the anomaly is a potential source. Velum scores each one to point investigators in the right direction.

• The ranking model uses three signals: how far upstream the facility is, what type of emissions they're licensed to produce, and how closely their profile matches the satellite observation.

• The output is a ranked list, not a guess. Each suspect comes with a confidence score and an estimated travel time for the discharge, with the most likely source rising to the top.

Team Members

• Ștefan Alexandrescu

         SWE @GARMIN, BSc in CS, AI Master's Student @UBB

• Iosif Blezu

        SWE @Raiffeisen Romania, BSc in CS @UBB, CyberSecurity Master's Student @UTCN

• Dragoș Buian

        Full-Stack Dev @Tapptitude, BSc in CS, AI Master's Student @UBB

• Alex Pavel

        DevOps Engineer Intern @Betfair, Math and CS Student @UBB

• Mihai Secoșan

       Freelancer, BSc in CS, Business Management Master's Stundent  @UBB

• Alexia Zaharie

       Freelancer, BSc in AI @University of Gronningen