Working with industry to build trustworthy AI for real-world impact

At the University of Waterloo's Critical Machine Learning (ML) Lab, Dr. Sirisha Rambhatla and her research team are working on building safe and efficient artificial intelligence (AI) models to advance sectors such as health care and aviation and help address climate challenges.

The team's research draws on foundational machine learning (ML) theory which not only tests models for reliability but goes deeper to predict and engineer it. In practice, this means users can make safer and fairer decisions at lower computing costs and with less waste.

Rambhatla, an assistant professor in the Department of Management Science and Engineering and the Val O'Donovan Chair in Efficient, Safe, and Adaptive AI, describes her work as "focusing on building adaptive and responsible models that perform well in the real world, not just in the lab."

"Whether its predicting flight delays or aiding organ transplant assessments, the AI acts as a support that enables humans to make faster, better decisions in critical situations," she says.

Research for the real-world

The lab's research projects and collaboration span several industries and partners.

In health care, for instance, the team collaborates with Dr. Mamatha Bhat, a professor at the University of Toronto and a clinician-scientist with University Health Network, Canada's largest research and teaching hospital network, to improve how clinicians match donor livers with transplant patients. Using AI, the lab is developing tools that can suggest the best match for a given recipient-potentially improving patient outcomes.

In the aviation sector, the lab works closely with Navblue, an Airbus company based in Waterloo, to help reduce flight delays. Their predictive models support smarter crew scheduling, reportedly improving delay performance by up to 60 per cent on the most uncertain travel days, ultimately impacting carbon emissions as well as personnel and customer experience.

The team collaborates with Apple on AI for Intelligent Production monitoring looking at how to adapt AI models trained in one environment to work reliably in another - with applications ranging from health care and manufacturing to autonomous driving.

"Imagine building an autonomous driving system in sunny California, then trying to use it in snowy Ontario," says Rambhatla. "The camera data is totally different. That's what we call a distribution shift - and we need AI that can handle it."

Supported by a NSERC Discovery Grant, the lab's research also looks at improving representation learning, training the ML models to "see" the world in separate concepts. For example, not automatically correlating big feet with a certain body size and shape, which can help generative AI modelling in health care deliver more reliable patient diagnoses.

AI & data research with Dr. Sirisha Rambhatla

Efficiency with impact

An important aspect of the lab's work is its climate-conscious approach to AI. Training large models, such as those used in training large language models (LLMS) like ChatGPT, can require vast computational resources - driving up both carbon emissions and costs.

"We're looking at how to train these models faster on modest graphical processing units (GPUs)," Rambhatla says. "Right now, building and using AI needs a lot of water for cooling, and the energy it uses for training contributes significantly to emissions. If we can do it faster, we can use far less energy in the process."

A recent algorithm developed in the lab has already shown promise in reducing LLM's training time by 43 per cent, while maintaining accuracy. Projects like this also feed into broader goals around AI equity and accessibility, ensuring smaller organizations with less resources can contribute meaningfully to generative AI without depending solely on a few companies with greater resources.

Industry collaboration and student opportunities

From algorithm design to testing to implementation, the Critical ML Lab does research that is relevant and useful.

"We don't just build models in isolation," says Rambhatla. "We co-develop them with our partners - clinicians, engineers and frontline workers - because that's how we make sure they actually work and can be used and trusted."

Dr. Sirisha Rambhatla with students from the Critical Machine Learning Lab.

This collaborative, grounded approach extends to the lab's internal culture as well. Graduate students, undergraduate research assistants and co-op researchers are all actively involved in shaping research questions and engaging with industry collaborators.

Chang Liu, a recently graduated master's student and recipient of Waterloo's 2025 Alumni Gold Medal, worked on Rambhatla's team and called it "meaningful work".

"Most self-driving systems trained on sunny weather data fail in Canadian winters," Liu says. "My work focused on improving these systems for snowy weather. I knew my contributions could help prevent accidents and save lives which made the research especially important."

The lab's emphasis on mentorship and practical impact makes it a compelling destination for students who want to push boundaries - not just in AI theory, but in ethical application and social responsibility.

Looking ahead, Rambhatla hopes to expand the lab's work on democratizing AI - making it possible for universities and even clinics to fine-tune powerful models without a supercomputer. "There's a big gap between who can use AI and who can afford to build it," she says. "If we can lower that barrier, we unlock safe and efficient innovation for everyone."

Dr. Rambhatla's work was recognized by the Faculty of Engineering with a 2025 Engineering Research Excellence Award.

If you'd like to find out about opportunities at the Critical Machine Learning Lab for graduate studies or industry partnership, please get in touch with Dr. Sirisha Rambhatla.

Feature image: Dr Sirisha Rambhatla (front centre) with her students from Waterloo's Critical Machine Learning Lab. Photo credit @ Sam Chen.