Traditionally the solution to the problem has been simply to aerate the water – using mechanical interventions such as pumps, diffusers and splashbox paddles. But water habitats are a fragile ecosystem – and can go from normal to hazardous very quickly. Activating the interventions is usually too little too late, as fish deaths tend to happen all at once. So the answer in commercial fish farms has been to run these machines 24/7, at immense cost and energy consumption.

A single commercial fish farm, for example one we know in Saudi Arabia, can run 5,000 aerators non-stop. Given that it costs £6,000 per aerator per year in energy bills, we could see clearly how unsustainable this approach to water quality management is.

We needed to build a full tech stack to monitor and predict both water quality and external (environmental, chemistry and weather) factors, in order to activate the right interventions at the right time. Crucially, given that this also means running certain machines very infrequently instead of permanently, we also needed to be able to remotely monitor, maintain and test them so they wouldn’t fail at the critical moment.

When we joined the project, we inherited a prototype that required a redesign of the hardware into a smaller, efficient, predictable, and production-ready device. We then built the supporting cloud infrastructure for ingesting and visualising real-time data, added multiple control modes (manual, threshold-based with hysteresis buffer, scheduled mode) and created a dashboard with user friendly controls for managing automated responses to environmental change: oxygen saturation levels falling below a threshold automatically switch on an oxygenator, plus an SMS and email notification system can be triggered.

We also integrated Victron VRM data for solar-powered installations, as many of the locations are remote and off-grid. Our tech stack comprised hardware, software, and firmware: beginning with an ESP32 microcontroller with MicroPython for device logic. Connectivity came in the form of a 4G cellular modem (with fallback to 2G/GPRS for resilience). We used modern Ethernet-abstracted modems with built-in APIs for signal strength/metadata, two devices supported Teltonika TRB246 and RUT906.

IoT Security was ensured by AWS IoT certificates for device authentication and encrypted comms, while firmware updates were handled securely via S3 with restricted upload tokens. Application security owed to Filament 4 dashboards with native 2FA and role/permission panels built-in. We avoided black-box third-party plugins, instead choosing an open-source base platform vetted by community.

The application layer was based on the latest Laravel and Filament php stack for APIs, taking advantage of Filament’s built-in panels for role-based dashboards and 2FA. Time-series data was stored in AWS exposed via dashboard and future app. The AI layer transforms vast amounts of raw environmental data into actionable predictions, risk alerts, optimisations, and planning insights. Cursor and GPT-5 / Sonnet 3.5 were used to accelerate coding of complex features like natural-language device status logic to deliver “Aeration Intelligence”.

Now proven in real-world usage, we have seen that this technology is immediately scalable not only in the relatively niche market of fisheries, but potentially globally for governments, environmental bodies, the academic sector and any organisations working in water management.

Until this technology was developed, there was no way of remotely monitoring, predicting and responding to water quality data in this way, so it is truly a world first. Within six months of deployment we had already captured 105 million data points, achieved more than 400% energy reduction, and zero fish deaths.