Data Engineering & Pipelines
A reporting team loses credibility the first time a dashboard shows yesterday's numbers during a board meeting. Behind almost every case of that is a pipeline that failed quietly, a load that partially completed, or a source system that changed a column without telling anyone. This programme covers building pipelines that move data reliably, notice when they have not, and can be re-run over a period without producing duplicates.
Programme Agenda
Pipeline Architecture
Sources, landing, staging and serving layers. ETL against ELT and why the warehouse doing the transformation changed the default answer. Batch, incremental and streaming, and choosing honestly between them.
Ingestion Patterns
Database extracts, file drops, API pulls and change data capture. Handling pagination, rate limits and an API that returns different shapes on different days. Landing raw data unchanged and why that matters later.
Incremental Loading
Full refresh against incremental, watermark columns, late-arriving records, and idempotent loads so a re-run does not double the numbers. Handling deletes in the source.
Transformation and Modelling
Modelling for analysis: facts, dimensions, grain, and slowly changing dimensions. Keeping transformation logic in version control and testable rather than buried in a scheduled query.
Orchestration and Scheduling
Dependency graphs, retries with backoff, timeouts, and what should happen when an upstream task fails. Avoiding the schedule that assumes an upstream system is always on time.
Data Quality Testing
Row counts, uniqueness, referential integrity, freshness and value range tests placed at pipeline boundaries. Deciding which failures should stop a load and which should raise a warning.
Backfills and Schema Change
Re-running a pipeline over a historical period without corrupting existing data. Absorbing an upstream schema change, and the contract conversation with the source system owner.
Observability and Cost
Logging, run history, alerting on failure and on suspiciously successful runs. Warehouse cost drivers, and the query patterns that quietly multiply a monthly bill.
Learning Outcomes:
Design a layered pipeline architecture with a clear responsibility per layer
Choose between batch, incremental and streaming for a given source
Build ingestion that survives API limits, pagination and shape changes
Implement idempotent incremental loads that handle late and deleted records
Model warehouse tables at a defined grain with slowly changing dimensions
Orchestrate dependencies with retries, timeouts and sensible failure behaviour
Place data quality tests at boundaries and decide what blocks a load
Run a backfill and absorb a schema change without corrupting history
Duration: 2 Days (16 Hours)
Training Hours: 9:00 AM to 5:00 PM
Level: All levels
Training Mode: Physical, Online, or Hybrid
HRD Corp SBL-KHAS Claimable
Certificate of Completion included
Frequently Asked Questions
More in Data Engineering and Machine Learning
- Data Science and Machine Learning Fundamentals · Beginner, 1 day
- MLOps and Model Deployment Basics · Beginner, 1 day
- Cloud Platform Foundations (AWS and Azure) · Beginner, 1 day