Where AI meets
the data platform.

Document review

Extracting and validating fields from registration documents, IDs and consent forms — with confidence scores that route low-certainty cases to a human.

Completeness checks

Flagging registrations missing documents, signatures or guardian details before they become a compliance problem mid-season.

Exception identification

Surfacing the records that don't fit the pattern — duplicate players, mismatched transfers, impossible dates — instead of waiting for someone to notice.

Administrative prioritisation

Ranking the day's follow-up queue by deadline risk and financial exposure, so limited admin time hits the highest-value items first.

Decision support

Summarising a player or club's full history into the three facts an administrator needs before making a call.

Documentation generation

Turning pipeline metadata and lineage into first-draft technical documentation that an engineer edits rather than writes.

Lab 01 — Registration document extraction

Testing structured extraction against real (anonymised) club registration packs. Measuring field-level accuracy versus manual entry, and the human review rate needed to beat the current error baseline.

• Document AI
• Human-in-loop
• Accuracy eval

Lab 02 — Natural-language pipeline queries

Can an administrator ask "which under-14s are missing a medical form?" and get a trustworthy answer over Unity Catalog? Focus is on refusal behaviour when the model shouldn't guess.

• Text-to-SQL
• Unity Catalog
• Guardrails

Lab 03 — Automated data quality rules

Generating candidate pipeline expectations from profiling a bronze table, then measuring how many survive review by an engineer who knows the domain.

• Profiling
• Expectations
• Lakeflow

Lab 04 — Migration assistant for M-code

Translating Power Query M-code into PySpark with a test harness that diffs output frames, so the translation is verified rather than trusted.

• Code translation
• PySpark
• Diff testing