Canlin ran on spreadsheets and internal tribal knowledge, with a diversified portfolio of thousands of wells whose data sat in SCADA, production accounting, reserves, and financial systems but were never interconnected. By way of my Master's degree I became the company's primary Engineer-turned-data hire, so there was no team to inherit and no precedent for what any of the architecture or solution was supposed to look like.
A conservative organization struggled mightily with accessing relevant information in a reasonable amount of time, and even though my motivations were to integrate predictive work into our operations, I had to build the foundation of trustworthy datasets and descriptive analytics with the intention to work upward from there. Year one was Tableau as the corporate reporting layer and data mart — 100+ data sources, 50 workbooks and 20 Prep flows published — and once people were genuinely using it I could make the case for a real platform underneath, with dbt Cloud and Snowflake as the single source of truth tying every system together. ML came third, on a home-grown ML-ops layer of Streamlit, Snowflake and DataRobot that was sized to what Canlin could actually operate rather than to what I would have liked to build.
- The dbt Cloud / Snowflake platform that became Canlin's foundation for all operational data, BI, and advanced analytics. Automation built on top of it saved 50+ hours per week across Operations.
- Production ML at portfolio scale — a modified XGBoost forecaster predicting the next 30 days of production for 8,000+ wells, inferring daily into corporate reporting, plus 2-year forecasts, gas-well anomaly detection for hydrates, liquid loading and mechanical failures, and a well-risk classifier trained on SME labels I crowdsourced through a Streamlit active-learning app.
- Analytics for the Integrated Remote Operating Centre (IROC — the company's remote operations hub): real-time well dashboards, and a networkx-based model of what a given facility outage would actually take down.
- The regulated annual reserves evaluation, owned end to end for five years. This is the externally-audited NI 51-101 valuation of every corporate asset, feeding straight into audited financial statements. Careful data and engineering work drove technical revisions adding +$730M (NPV10) over four years, and that was through a commodity-price trough. The same rigor fed the A&D data support that helped take corporate debt from $120M to $0.
- Mentored junior professionals into data roles, and kept making the business case for analytics investment at the leadership table for seven years.
By the time I left, the people I'd mentored were running the operational reporting, the ML forecasting and the annual valuation on platforms I'd designed for them. Two years later most of the primary analytics continue to run with minimal oversight.
This is the data platform engineering pillar with ML and data leadership running underneath it.