Selected projects

Work built for consequential decisions.

Four operating environments. The common thread is difficult data, domain-heavy logic and systems that must remain trustworthy after the first successful run.

01

Industry leadership · Ballard Power Systems · 2021—present

From raw fleet telemetry to trusted operating intelligence.

Built and led the data foundation used to analyze a global fleet of fuel-cell buses and modules—connecting hundreds of telemetry signals to engineering KPIs, reliability analysis and fleet decisions.

  • Built the daily ingestion system solo, end to end. The Databricks package tracks per-module high-water marks, metadata and processing state across changing datasets.
  • Made failure and reprocessing safe. Idempotent reruns, staged writes, two-day backups, health checks and source-to-target reconciliation prevent duplicate or partial loads.
  • Handled late and corrected records. Targeted backfills recalculate a safe time margin and propagate affected outputs; complete module history can be replayed when dependencies demand it.
  • Turned telemetry into usable products. A scalable framework covers 20+ product types, 500+ tags and 50+ KPIs, with 15+ Power BI dashboards supporting engineering and management.
  • Extended the platform into applied AI. Work includes anomaly detection, forecasting, failure prediction and a Databricks-based RAG assistant for technical knowledge.
Fuel-cell electric transit bus outside Ballard Power Systems
Fuel-cell electric bus at Ballard’s Burnaby facility. Image: Ballard Power Systems.
5,000+fleet modules
20+product types
500+telemetry tags
50+engineered KPIs
02

Consulting engagement · BC Hydro · 2017—present

Replacing fragile reliability analysis with an auditable planning platform.

For BC Hydro Energy Planning, developed SHARE: a reliability and resource-adequacy platform that turns complex uncertainty into repeatable, reviewable planning evidence.

  • Rebuilt a legacy analytical workflow as a maintainable Python and SQL system with explicit inputs, assumptions and traceable outputs.
  • Modelled correlated demand, resource availability and system states using Monte Carlo methods, Markov processes and copulas.
  • Created validation, scenario comparison and reporting layers so planners can explain how each result was produced.
  • Supported a continuous consulting relationship spanning Energy Planning and selected Generation System Operations work.
Why it matters

Reliability models inform long-horizon resource decisions. The platform makes those decisions easier to reproduce, challenge and defend.

03

Consulting via Theory and Practice · Infrastructure Ontario · 2020—2021

Accelerated High-Speed Internet Program (AHSIP)

Spatial intelligence for a province-wide reverse auction.

Built the geospatial data and feature foundation used to divide Ontario into balanced broadband bidding lots—so providers could compete for subsidies across defined service areas.

  • Collected and reconciled public census, development, settlement, connectivity and infrastructure datasets from government sources.
  • Designed PostGIS schemas and dbt transformations, then wrote SQL feature pipelines for hexagonal spatial blocks.
  • Combined connectivity and development features into a representative score that could be profiled and used by clustering and optimization methods.
  • Helped design the lot-forming strategy, including spatial clustering and score-based seed selection; the core optimization implementation was written by the project’s lead data scientist.
  • Took over model execution after the lead’s departure, corrected defects, reran calculations and supervised three junior analysts on exploratory and query work.
View Infrastructure Ontario’s AHSIP page
Fibre-optic strands representing high-speed internet infrastructure
Program image: Infrastructure Ontario.
Ontario public map showing a funded AHSIP project area
Public Ontario High-Speed Internet Projects & Availability map.

Public program context

Infrastructure Ontario reports that AHSIP’s reverse-auction process awarded contracts to eight internet service providers, covering up to 266,000 unserved and underserved homes and businesses across as many as 339 Ontario municipalities. These are program-level outcomes, included to establish the scale of the initiative.

04

Consulting via Theory and Practice · G&F Financial Group · 2020—2021

Trustworthy customer analytics from difficult monthly data.

Created the ingestion and analytical workflow for masked financial-customer records, then used the resulting longitudinal data to identify patterns, segments and anomalies.

  • Built Python ingestion for monthly SQL Server and CSV deliveries into PostgreSQL, managing schema differences and difficult file behaviour.
  • Added load metadata, count checks and source-to-target reconciliation so every monthly refresh could be verified.
  • Reconciled panel data across periods and applied regression, logistic regression, clustering, segmentation and anomaly detection.
  • Converted one-off analytical work into a repeatable process suitable for new customer records each month.
Panel reconciliationLogistic regressionSegmentationAnomaly detectionPostgreSQL

Have a related problem?

Start with the system, the data and the decision.

AP Analytics can help diagnose the path from unreliable inputs to a production-ready analytical system.

Discuss a project