Bell Canada · Data Engineer
June 2025 – PresentI own 25+ production data pipelines, including 14 scheduled control flows, behind field-operations and executive reporting.
$26.8Mmisreported order value corrected
28K+dropped records recovered
3.5Mduplicate rows removed
28K+ records, silently dropped. then recovered.
- Own 25+ production data pipelines, including 14 scheduled control flows, behind field-operations and executive reporting: monitoring, incident response and change control; mentor teammates on debugging and root-cause analysis.
- Corrected $26.8M of misreported order value in a $40M+/year pipeline: restored $4.1M under-counted (including $2.7M across 8,355 missing orders) and removed $22.7M double-counted, through a reversible production cutover with zero duplicate-key groups.
- Owned a P1 incident on an executive KPI dashboard: reproduced it cell for cell from the fact table and isolated two opposite ETL defects across 13 sources; both fixes validated (mangled IDs from 13,445 to 0).
- Built the 3-tier nightly batch ETL/ELT pipeline (staging, warehouse, analysis) on Teradata and SAS DI behind network-ticket reporting, integrating four systems into a 78-column fact table; 150K records/day since November 2025; presented to directors.
- Technical lead for the team's first MongoDB-to-BigQuery migration pipeline on the GCP data platform: designed the scalable landing schema (partitioned, clustered keys plus a JSON payload; partition pruning for query cost optimization).
- Wrote the pipeline's Terraform, Cloud Composer orchestration (DAGs) and Cloud Run ingestion jobs via GitLab CI/CD, with watermark-based change data capture (CDC) and MERGE upserts.
Show the full role record (12 more)Hide the extra 12 entries
- Set technical direction: authored the approved high-level data architecture for a four-layer BigQuery data warehouse (landing, core data models with declared grains and schema-contract tests, data marts, semantic layer) and designed its nested-order unpacking, with a PySpark-versus-BigQuery SQL pilot planned.
- Ran a large-scale backfill of 42M+ documents from 5 production MongoDB Atlas collections into a BigQuery dev dataset; caught a silent half-load by reconciling source counts, and fixed Airflow zombie tasks and out-of-memory batches.
- Built and offline-tested a reliability hardening package for 5 Airflow DAGs (the watermark advances only after a successful MERGE, pre-merge validation, empty-batch no-ops): 16 behavioral tests and 42 static checks, now in code review.
- Built a Python reconciliation toolset (value diff, ID-set diff, checkpoint-resumable row hashing via the BigQuery Storage Read API) and compared 6.09M rows across 5 tables on QA-sourced development data: 99.98% identical.
- Cut a 23M-row join 83% (12 to 2 min) through performance tuning (filtered lookups, predicate pushdown, skew analysis) before go-live; traced a $10M gap in internal revenue reporting to one date filter.
- Recovered 28K+ silently dropped records and grew usable history from 1 to 9+ months; removed 3.5M duplicate hierarchy rows with point-in-time SCD Type 2 joins.
- Replaced a 1,000-line legacy SQL view with a 57-step SAS DI ETL (complex joins across 13 source tables) building order- and circuit-grain fact tables, surfacing 74 orders the old view dropped.
- Took an automation-rate KPI from business requirements to build: verified the business rule across 3.39M interactions, rejected a defective draft SQL, and added the measure across 7 SAS DI jobs.
- Root-caused missing billing addresses across 6 billing systems and won written business approval for the fix; accounts with a usable province rose 6.4x (147K to 947K).
- Analytics engineering and BI: built a MicroStrategy Intelligent Cube (77 attributes, 12 metrics) for self-serve reporting, plus KPI dashboard data products released Dev to Prod; designed data quality alerting (freshness and volume checks, rolling P10/P90 thresholds).
- Named change-control gatekeeper and peer reviewer 7 weeks in; facilitated weekly cross-functional stakeholder meetings; closed 157 of 170 tickets (92%, 72% P1/P2); set data governance and documentation standards (data dictionary, data lineage, metadata management, view-based access isolation for external accounts).
- Use AI coding assistants (Claude, Gemini, ChatGPT) for pipeline scaffolding and SQL review; designed a Teradata history layer for LLM-generated quality scores (idempotent reloads, retention).