Kirsti Racine entered the world of AI well before it became buzzy.
Long before becoming VP, AI Technology Lead at TD, she had completed a master’s thesis in AI at Simon Fraser University, and worked at IBM, where she helped develop early machine learning systems for healthcare, focused on speeding up the process of routing patients to doctors for MRIs.
“I was told I’d never have a career in AI, because it wasn’t all that popular back in the day,” Racine said.
“We spent a long time and a lot of calories trying to help our colleagues feel very comfortable and very confident with what they’re doing from an AI perspective.”
Kirsti Racine, TD
Today, she heads up the team at the centre of an ambitious push for AI adoption across one of Canada’s largest banks. The platform she oversees, known as the TD AI Platform (TAP), serves as the central hub for TD AI infrastructure.
It supports every line of business, from banking to wealth management, insurance, marketing, and more.
TAP is designed to make AI practical and accessible to both engineers and everyday employees.
It underpins most of the bank’s most significant AI deployments to date, and according to Racine, it’s already enabling teams across the bank to solve problems and streamline tasks in ways they hadn’t anticipated.
“Some of the things that people have managed to do with the tooling that we’ve provided have been unexpected and delightful,” Racine said. “Just one example is how our finance team started using the platform to build out use cases specifically designed to explain financial data attributes to employees in plain English. They have now developed a robust set of use cases which they’re deploying across their function.”
From pilots to platforms
The origins of TAP trace back to late 2022, just as generative AI was gaining traction globally. TD had already invested in AI, acquiring Toronto-based AI firm Layer 6 in 2018 and completing a major cloud data platform migration by mid-2024.
“We got a bit lucky at TD,” Racine said. “We took in the machine learning ops engineering pods from the Layer 6 team, and we took in senior talent and leadership from our cloud data platform, and that formed the nucleus of what became TAP,” Racine added.
Within months, TD launched two early generative AI tools. One was a deep-search assistant that helps contact centre agents quickly find relevant policies and procedures, which was built in close collaboration with agents.
Another initiative completed last year was launching a pilot for GitHub Copilot, an AI programming assistant, across its developer ecosystem. This capability analyzes engineers’ code and provides real-time suggestions for completing and testing. The rollout was paired with comprehensive training and ongoing support to engineers, and today, 92 percent of TD engineers engage with Copilot on a weekly basis.
“We spent a long time and a lot of calories trying to help our colleagues feel very comfortable and very confident with what they’re doing from an AI perspective,” Racine added.
Building the foundation
TAP now encompasses the foundational layer that allows AI to be deployed across TD in a secure and repeatable way. According to Racine, the team behind it is lean, just 60 people, but its impact touches nearly every corner of the organization.
TD currently runs AI environments, tailored to each business line, which support both predictive and generative AI use cases. Everything from monitoring to bias detection to deployment workflows is automated, which has cut down the time and cost of deployment by as much as 85 percent.
One of the platform’s most technically ambitious projects is TD AI Prism, a predictive foundation model built to allow engineers to develop predictive models with greater accuracy and speed. It is also the largest AI model ever deployed by TD, made possible by TAP’s fully automated production deployment capabilities.
“TD AI Prism, to me, is a differentiator,” Racine said. “It’s one of the things that TD has done that I have not seen anyone else do. We’ve really transformed the way that we’re doing predictive AI, and this enables us to deploy models at scale that we’ve never been able to see before.”
Fast traction
TAP was built by the bank’s internal cloud and data teams and the machine learning professionals from Layer 6, and according to Racine, that relationship remains central to the platform’s evolution. Layer 6 provides many of the models, while Racine’s team focuses on machine learning ops, infrastructure, and deployment frameworks.
Both Racine and Maksims Volkovs, Chief AI Scientist at TD and Co-founder of Layer 6, credit that working relationship for allowing each business unit to build and own their own AI applications, using shared infrastructure and guardrails, without slowing teams down.
“Kirsti and I share the same view that AI is the key to unlocking next generation solutions, not only for our clients’ benefit, but for our employees and businesses within the bank as well. We aim to enable AI at scale, and we aim to enable AI at the edge of this organization, so that every line of business has direct access to this cutting-edge technology,” Volkovs added.
Following the launch of the contact centre assistant, teams from all business lines began working on their own generative AI use cases. Each of them used TAP’s components and frameworks to get started faster, which Racine described as “a true moment of satisfaction” that what her team was building was gaining traction across TD.
AI that “just works”
The 2025 TD AI Insights Report revealed that 64 percent of respondents don’t feel they’ve received adequate training on AI at work. It’s a pain point that helped shape the bank’s approach to TAP.
Training is built into the deployment strategy, delivered in the format that works best for each employee, whether virtual, in-person, or on-demand. This approach has led to higher adoption and stronger engagement across the board at TD.
The team also continues to invest in the soft skills that support technical transformation, including change management, collaboration across teams including infrastructure engineering, and user-centred design. Every deployment is treated as a product with real user feedback and continuous iteration.
Racine’s team continues to expand TAP’s reach, with more environments, more automation, and new models under development. While most current applications focus on internal operations, she said client-facing generative AI tools are on the roadmap for 2026.
Racine’s vision is of AI as infrastructure that’s deeply embedded, dependable, and largely invisible. That perspective shapes how she hopes TD employees will come to regard it in the years ahead.
“I’m hoping that they’ll say AI at TD just works,” Racine said. “That it’s intuitive, it’s secure, and it helps our people do their jobs better every day.”
PRESENTED BY
At TD, we’re building AI systems that are intuitive, secure, and grounded in real-world needs. Read more about TD AI Prism.
All photos provided by TD.