Case Studies

The work speaks for itself.

Every engagement starts with a business problem and ends with a delivered outcome. Here is a selection of our past work.

Retail & E-commerceAgentic Analytics & Predictive IntelligenceProduction system — ongoing

Tariff impact analysis identified Shopify merchants at risk of inventory stockout

The Challenge

Shopify serves millions of merchants and needed a way to identify which ones were at risk of inventory stockout due to supply chain disruptions. As tariff changes began rippling through global trade flows, the platform needed to surface at-risk merchants proactively.

Our Approach

Designed and built a 7-model dbt pipeline that combined inventory levels, order velocity, and external disruption signals into a structured, auditable data layer. The pipeline fed downstream intelligent algorithms that scored merchants by stockout risk and surfaced actionable signals. Deployed internally to monitor the real-time impact of tariff changes on merchant inventory health across the platform.

Outcome

The pipeline became the authoritative internal source for merchant inventory risk during periods of supply chain volatility. Teams could identify at-risk merchants early, enabling proactive outreach and informing platform-level responses to tariff-driven disruptions.

dbt Models

Production

Deployment

Tariff impact

Use Case

Auditable

Data Layer

Tech Stack:PythonBigQuerydbtAirflowSQLGoogle Data StudioGoogle Cloud Platform

Retail & E-commerceAgentic Analytics & Predictive IntelligenceInternal framework — adopted org-wide

Back office segmentation gave Shopify a shared language for understanding merchant complexity

The Challenge

A product team operating across a large commerce back office had no structured view of how merchants actually ran their operations. Third-party app usage varied enormously across the merchant base, but without a common framework, product decisions, merchant interviews, and feature targets were all made without accounting for that complexity.

Our Approach

Extracted third-party app functionality signals from multiple data sources. Apps were automatically categorized into complexity tiers using a classification pipeline, then aggregated at the merchant level to engineer a five-tier Back Office complexity segmentation. Each tier represented a distinct merchant profile based on the sophistication of their operations stack.

Outcome

The five tiers became the lingua franca across product, design, and research — used to recruit the right merchants for interviews, to set differentiated adoption targets for new feature launches, and to bring systems thinking to a product area that had previously reasoned about merchants as a monolith.

Complexity Tiers

Multi

Data Sources

Team-wide

Adoption

Roadmap & research

Impact

Tech Stack:PythonBigQuerydbtSQLLooker StudioGumloop AutomationGoogle Cloud Platform

Government & DefenseAgentic Analytics & Predictive Intelligence6-year research program (2012–2018)

Unsupervised anomaly detection system strengthened maritime domain awareness for the Canadian Navy

The Challenge

Defence Research and Development Canada (DRDC) and the Canadian Navy needed to detect anomalous vessel behavior across vast maritime areas. Sensor data from AIS transponders, GPS, LiDAR, and satellite imagery arrived in incompatible formats at different cadences — creating a fragmented operating picture that made systematic anomaly detection impractical.

Our Approach

Designed and built an unsupervised learning system that inferred normal maritime traffic lane distributions from historical GPS and radar data, then flagged statistically significant deviations as anomalies requiring human review. Developed multi-sensor fusion algorithms to combine AIS, GPS, LiDAR, and satellite imagery into a coherent maritime common operating picture. The system was developed by Larus Technologies in close consultation with DRDC Ottawa and Valcartier over a six-year research program.

Outcome

Larus Technologies' operational maritime domain awareness system was deployed with the Canadian Navy. Research contributed to multiple peer-reviewed publications on high-level information fusion for defense applications. Methods developed were adopted into ongoing Canadian maritime security programs.

Sensor Types Fused

6 years

Research Program

Operational

Deployment

Peer-reviewed

Publications

Tech Stack:MATLABRGISAIS/GPS sensorsLiDARSatellite imagery

Technology & SaaSAgentic Analytics & Predictive IntelligenceInternal tooling — ongoing

Agentic analytics system synthesized a metric hierarchy and automatically diagnosed anomaly drivers for Shopify

The Challenge

A data science team supporting Shopify's commerce back office had no formal framework connecting individual product initiatives to top-line outcomes. Without a structured metric hierarchy, anomaly investigation was manual and time-consuming — each time a key metric moved, the team had to trace drivers from scratch with no systematic starting point.

Our Approach

Built an AI-powered analytics system that synthesized a metric ladder — north star, domain, and driver metrics — by combining existing product roadmap initiatives with qualitative merchant feedback. Developed agentic skills that continuously monitored the hierarchy, automatically detected metric anomalies, identified the drivers behind each movement, and surfaced recommended product actions. The system reduced the cycle from anomaly to actionable insight from days to minutes.

Outcome

Adopted by the full 10-person data science team as the shared framework for product health monitoring.

10 people

Team Adoption

3 levels

Metric Tiers

Days → mins

Anomaly to Insight

Roadmap + feedback

Data Sources

Tech Stack:PythonClaudeBigQuerydbtSQLCursorGoogle Cloud Platform

Ready to write your own case study?

Let's map the highest-value AI and data opportunity in your business.