Transforming water management with predictive AI on AWS with Ainwater

Ainwater’s solutions impact the lives and water quality of more than 1 million people in Chile, Mexico, and Brazil. The company harnesses AI to help cities and companies run their water treatment, desalination, wastewater, and drinking water plants better. “Water doesn’t wait, nor does the planet,” says Camilo Huneeus, CEO of Ainwater. “That’s why we need to build solutions that facilitate water security. That includes public policy, engineering, and software and AI for making sure facilities run properly.”

Ainwater built its solution using machine learning and large language models (LLMs) to power predictive insights. It then needed a secure, compliant environment that it could use to cost-effectively scale out its solution across multiple regions. After testing another cloud provider, Ainwater chose AWS and was able to smoothly integrate its Python and Django workflows into an AWS environment, reducing the complexity of its architecture. “Using AWS, I don’t have to buy a supercomputer to train a model,” says Dr. Marcos Pérez, CTO of Ainwater. “The opportunity to access massive compute and data processing power on AWS has increased our ability to train and deploy models.”

Ainwater’s multi-tenant SaaS solution, Poseidón, delivers scalable, data-driven water management to industrial and utility clients across Latin America using AWS. With this solution, Ainwater is changing the paradigm from reactive management to predictive management using AI-driven insights and recommendations. “Thanks to the architecture we’ve built on AWS, we’re able to set up a system for running, controlling, and predicting how a plant is going to behave—and to do so in a way that’s easy and quick for us and for the client,” says Huneeus.

The solution helps customers reach their sustainability goals by providing plant operators and managers with operational suggestions on how to run drinking, desalination, and wastewater treatment plants. For example, if Ainwater’s models predict that the effluent contamination might breach its limit, the solution provides specific operative recommendations to stay compliant. These predictions help customers save time, money, and energy by running their processes more efficiently.

Poseidón is designed with a modular architecture that integrates near real-time monitoring, predictive analytics, digital logbooks, reporting, and even computer vision for wastewater microscopy. The solution ingests heterogenous industrial data from various sources and turns it into actionable recommendations for customers using machine learning models, such as time series forecasting or models to optimize chemical dosing. Ainwater embeds these predictive models directly into its software using Amazon SageMaker AI.

To scale out its solution to multinational clients, Ainwater implemented AWS Fargate, serverless compute for containers. “For the scalability and stability of the solution, we use AWS Fargate as the core,” says Pérez. Ainwater uses AWS Fargate to offload infrastructure management for compute instances, storage buckets, APIs, and network configurations. With this architecture, implemented using Terraform, Ainwater could expand its solutions to clients across multiple countries, using local AWS Regions to keep latency low.

The company also uses Amazon Bedrock—a comprehensive, secure, and flexible service for building generative AI applications and agents—to experiment with new LLMs. The service offers implementation and architecture in one place, and Ainwater benefits from having access to the latest models.

On AWS, Poseidón 3.0 has scaled out to over 100 plants in 3 countries. Client onboarding is 65 percent faster than the initial version, and model development is 80 percent faster using Amazon SageMaker AI and Amazon Bedrock. Additionally, Ainwater’s delivery speed increased by 30–40 percent compared to its previous pipelines. “Using AWS, we’ve been able to continuously improve how we work,” says Huneeus. “We’ve been able to integrate data much faster and with less human configuration using Amazon SageMaker AI frameworks alongside the architecture and processes that we’ve implemented on AWS.”

By using AWS, Ainwater also saves energy, which the company can determine using the Customer Carbon Footprint Tool to track, measure, and review the carbon emissions generated from AWS use. “On AWS, we’re able to run models efficiently, which means using fewer energy resources and therefore less CO2 emissions,” says Huneeus.

The solution helps Ainwater’s customer organizations provide better water treatment to 1 million people. By using predictive models, Poseidón helps customers take proactive measures—in wastewater, for example, adjusting processes based on expected biological parameters like bacteria growth—which lets them use fewer chemicals, save time, and reduce costs. For one customer in the meat packing industry, Ainwater created a model to predict oxygen levels in the aeration process of water treatment at the plant, and its insights delivered 9 percent energy savings for the customer. The solution has led to 35 percent reduction in grey water from industries—equating to more than 3 million cubic meters per year—as well as a 30 percent energy reduction at water treatment plants, 10 percent reduction in chemicals for water treatment, and 82 percent more water produced by desalination membranes.

Ainwater will be continuing to expand its customer base and develop its technology, particularly in agentic AI. “We will be the first in the world to deploy agentic AI for water treatment, and to do this, we need a backbone of top-notch cloud services,” says Huneeus. “By using AWS, we can develop and deploy these new technologies that will become the future standard for water treatment, helping to improve water quality across the globe.”