Amii 擴大 AI 應用合作計劃，支援另外 25 間企業

Amii, the Alberta Machine Intelligence Institute, is expanding its partnership with Canada’s regional economic development agency PrairiesCan to help 25 more small and medium-sized businesses develop practical AI capabilities. The October 27 announcement adds C$2.5 million in federal support, bringing the initiative’s total federal investment to C$5.6 million.

The expansion covers Alberta, Saskatchewan, and Manitoba, with initial engagements underway in energy, agriculture, and manufacturing. Amii says the program has worked with 30 regional businesses since launching in 2024. Its approach pairs companies with machine-learning specialists to build systems around their own operational data and leave them with the expertise to maintain those systems.

A Research Partnership Built Around Implementation

Based in Edmonton, Amii is an independent nonprofit and one of Canada’s three national AI institutes under the Pan-Canadian AI Strategy. Its history stretches back to the Alberta Ingenuity Centre for Machine Learning in 2002; it became an independent nonprofit under the Amii name in 2017. Its work connects research, talent development, and commercial applications.

Prairies Economic Development Canada, known as PrairiesCan, is the federal agency supporting economic development in the three participating provinces. For international readers, the partnership is an example of public investment helping smaller companies access specialist technical capacity that would be difficult to assemble independently.

Amii’s AI development model places a full-time machine-learning resident on a company’s project, supervised by staff scientists. The resident works with the business team on model development and testing, with an option for the company to hire that person afterward. The general offering lists a 12-to-15-month commitment, although individual project scope and cost vary.

Companies retain the intellectual property developed through the engagement and receive documentation and knowledge transfer. That gives the partnership a different objective from simply delivering an outsourced prototype: the business should be able to adapt the system after the external research team leaves.

Croptimistic: Computer Vision That Can Handle Real Fields

Croptimistic Technology develops the SWAT MAPS precision-agriculture ecosystem, which combines soil, water, and topography information to guide field management. Its SWAT CAM product collects images during normal sprayer operations and uses machine learning to analyze crops and weeds.

The Amii collaboration tackled early-season cereal crops such as wheat and oats. Their narrow, overlapping leaves make individual plants difficult to distinguish. The team developed specialized vision models to work through dense canopies and challenging imaging conditions, including dust, glare, and wind.

Croptimistic’s product information explains how sprayer-mounted cameras connect dense field imagery to management zones. The practical value of better crop detection is spatial detail: growers can examine uneven establishment and weed pressure across a field rather than relying only on a few manual observations.

Carson Integrated: Turning Environmental Records Into Predictions

Carson Integrated is an environmental and forestry consultancy whose work includes geospatial information for natural-resource projects. Its challenge was stream classification, which helps determine protective buffer zones around waterways.

Amii’s Carson case study describes a model built using roughly two decades of collected data, including field measurements and remote-sensing information. Feature development examined terrain and environmental characteristics; catchment outlet size proved particularly useful. The study reports average classification accuracy of approximately 75% when the appropriate data were available.

The application illustrates how domain expertise shapes useful machine learning. Identifying a waterway in imagery is a different task from predicting its regulatory classification. Predictions can help prioritize assessment work in remote terrain, while their accuracy and suitability still need to be evaluated for the decisions they support.

Visionstate: Scheduling as a Constrained Optimization Problem

Visionstate develops software for facilities operations, inspections, and work verification. Its current website emphasizes MIRA, a platform connecting reports, inspections, work orders, and oversight. The Amii engagement focused on the company’s WANDA facilities-management platform.

The detailed case study explains that cleaning routes and schedules were formulated as a constrained optimization problem. The system must allocate a fixed workforce while respecting site layouts, service requirements, and compliance rules, balancing efficiency with fairness.

The work also supported TidyLogic, a resource-planning product, and a healthcare cleaning-cost estimation tool. The case study described TidyLogic in beta and the model moving toward integration at the time of writing. These distinctions matter: commercial product development and a fully deployed optimization system are separate milestones.

Ambyint: Building Reinforcement-Learning Expertise

Ambyint supplies production-optimization software for oil and gas operators. Its existing platform combines physics-informed models with automated control, including adjustments to equipment setpoints through industrial monitoring infrastructure.

The Amii project explored reinforcement learning for plunger-lift wells, where changing pressure thresholds affect how accumulated liquids are removed. Sparse knowledge of conditions deep underground makes simulation and control difficult. The team tested historical-data approaches and physics-informed methods to improve decisions.

The case study is explicit about the outcome: the experimental models did not improve efficiency over Ambyint’s existing automation and were not yet reliable or responsive enough for production. The benefit was a better-equipped internal team and a clearer understanding of where further research was needed. That is a useful adoption result even when it does not immediately produce a deployable model.

Riva: Making Enterprise Data Useful for AI

Amii also identifies Edmonton-based Riva as an early beneficiary, citing progress toward production-ready solutions without disclosing the specific model architecture.

Riva’s platform connects email, calendar, contact, and meeting information with customer relationship management systems. Its capabilities include AI-generated conversation summaries and structured meeting outputs that teams can review and connect to CRM records. This provides a different adoption context from field imaging or industrial control: the challenge is making business interactions usable while preserving governance and human review.

Why the Model Matters Beyond Canada

The participating businesses demonstrate that AI adoption spans computer vision, predictive modeling, constrained optimization, reinforcement learning, and enterprise information workflows. Each requires different data, evaluation methods, and operational expertise.

The partnership’s wider significance is its emphasis on ownership and technical capacity. A company gains more than a model when its own staff learn how that model was built, how to test it, and when its predictions should be trusted. For smaller businesses, transferring those capabilities can be as consequential as the initial software delivery.