For data leaders and senior engineers accountable for the data behind AI systems: enforce quality with contracts and tests, trace lineage from source to model, document datasets, and meet EU AI Act, GDPR and copyright-driven data obligations.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
When a model misbehaves, the first questions are about data: what it was trained or grounded on, where that came from, who approved it and whether it was fit for purpose. This course builds the capabilities to answer them. It starts with quality dimensions that matter for AI beyond classic completeness and validity: representativeness, label quality, freshness for retrieval, leakage and contamination. You will encode them as data contracts and automated checks with Great Expectations, Soda and dbt tests, wire them into pipelines as blocking gates, and monitor drift in production data.
The lineage modules use OpenLineage events and catalogs such as DataHub, Unity Catalog and Apache Atlas to trace data from source systems through feature and embedding pipelines to model versions, so you can answer impact and deletion questions. Documentation modules cover datasheets for datasets, data cards and MLCommons Croissant metadata. The regulatory module covers EU AI Act Article 10 data governance for high-risk systems, the Commission's template for the public summary of training content for general-purpose AI models, GDPR lawful basis and erasure requests against trained models and vector stores, and text-and-data-mining opt-outs, treated as EU rules that reach US companies selling AI systems into the EU or processing EU residents' data. Regulatory content is educational, not legal advice. The course ends in an automated exam; passing earns a Wirewalk Institute of AI certificate of completion and assessment.
What you will be able to do
Define AI-specific data quality dimensions for a use case, including representativeness, label quality, freshness and contamination
Write data contracts and automated quality checks with Great Expectations, Soda or dbt and enforce them as pipeline gates
Capture dataset and column-level lineage with OpenLineage and query it in a catalog to answer impact and provenance questions
Trace a model version back to the exact datasets, versions and transformations that produced it
Produce datasheets, data cards and Croissant metadata that meet internal and external documentation needs
Map EU AI Act Article 10, the GPAI training-content summary, GDPR and copyright opt-outs to concrete data controls
Design a process to honor deletion and objection requests across training sets, fine-tuned models and vector stores
Syllabus
6 modules · 20 lessons · about 18 hours
1
Module 1: Data Quality for AI
Quality dimensions that predict model behavior, not just pipeline health.3 lessons
Beyond Completeness and Validity: Representativeness, Label Quality and Freshness
Leakage, Contamination and Train-Test Hygiene as Quality Rules
Bias and Coverage Audits of Training and Evaluation Data
2
Module 2: Contracts, Tests and Monitoring
Turning quality expectations into code that blocks bad data.3 lessons
Data Contracts Between Producers and AI Teams
Great Expectations, Soda and dbt Tests as Pipeline Gates
Monitoring Drift and Anomalies in Production Data and Embeddings
3
Module 3: Lineage from Source to Model
Knowing where every model input came from.4 lessons
OpenLineage: Runs, Jobs, Datasets and Facets
Catalogs Compared: DataHub, Unity Catalog and Apache Atlas
Connecting Data Lineage to Model Registries and Experiment Tracking
Impact Analysis: What Breaks, and Which Models Are Affected
4
Module 4: Documenting Datasets
Documentation that serves engineers, auditors and regulators.3 lessons
Datasheets for Datasets and Data Cards
MLCommons Croissant Metadata and Machine-Readable Documentation
Provenance and License Records for Third-Party and Scraped Data
5
Module 5: Regulatory Data Obligations
What the law requires of AI data, and the controls that satisfy it.4 lessons
EU AI Act Article 10: Data Governance for High-Risk Systems
General-Purpose AI: The Training-Content Summary Template and Copyright Policy
GDPR for AI Data: Lawful Basis, Minimization and Erasure Across Models and Vector Stores
Text-and-Data-Mining Opt-Outs and Honoring Them in Pipelines
6
Module 6: Operating a Data Governance Program for AI
Roles, workflows and metrics that keep governance working.3 lessons
Ownership, Stewardship and Approval Workflows for AI Datasets
Access Control, Retention and Audit Trails
Governance Metrics and Evidence Packs for Internal Audit
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 25 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 60 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by Great Expectations, Soda, dbt, OpenLineage, DataHub, Databricks Unity Catalog, Apache Atlas, or MLCommons. Product and company names are trademarks of their respective owners.
How data pipelines for AI differ from analytics pipelines, and how to build them: lakehouse storage with Parquet and Iceberg or Delta, distributed processing with Spark, Ray Data and DuckDB, text cleaning, filtering and MinHash deduplication at corpus scale, dataset versioning, orchestration with Airflow or Dagster, and feeding GPUs efficiently at training time.
Build production pipelines for unstructured documents: format detection and text extraction, layout analysis and reading order, OCR and vision-language parsers, table and form extraction, schema-driven field extraction with LLMs, human review loops, and running it all at scale with measured accuracy and cost.
Run annotation as an operation with measurable quality: task design and guidelines, labeling tools such as Label Studio, CVAT and Argilla, in-house versus vendor workforces, inter-annotator agreement, gold questions and adjudication, model-assisted and active-learning workflows, and the human preference and rubric labeling used for LLM post-training and evaluation.