TankFlare

Most aquarium mistakes are not made in the tank. They are made in the store aisle, minutes before checkout.

A beginner picks fish because they look good together, plants because they are green, and decor because it fits the theme. What they cannot see is that one fish needs soft acidic water while another needs hard alkaline water, that a plant needs more light than the tank will ever get, or that a beautiful piece of driftwood will slowly shift the pH in the wrong direction.

Compatibility information exists, but it is scattered across forums, care sheets, and contradictory blog posts. TankFlare pulls that reasoning into one place and applies it to the specific combination a person is actually considering.

The user describes their tank: volume, water parameters, lighting, substrate, and current inhabitants, then adds what they are thinking of buying. TankFlare evaluates the combination on three fronts.

Instead of returning a bare verdict, the AI explains why a pairing fails and what to change: a different species, a larger tank, or an adjustment to the setup.

TankFlare uses an ensemble model to turn aquarium details into a clear compatibility recommendation.

More than 15,000 data points were collected and organized using personal freshwater-aquarium knowledge. Specialist XGBoost and LightGBM models evaluate signals from the tank setup, water parameters, bioload, and the fish-and-decor combination. Each model contributes an independent opinion; a logistic regression model then combines those predictions into the final match result.

TankFlare taught me that the interesting part of an AI product is rarely the model. It is deciding what the model is allowed to say, what data it must reason from, and how to present an answer that a beginner will actually trust and act on.

It brought together my interest in aquariums and software development to help fishkeepers make better-informed decisions for the health of their fish.