Enterprise
Finding the failure inside 18M+ records
Enterprise data investigation across customer-data workflows at Deloitte USI: SQL-led root-cause analysis, earlier exception checks in Python, and the runbooks that made fixes repeatable.
Context
When data-quality issues show up late in a workflow, finding the source takes longer and downstream teams trust the output less. The question was whether exceptions could surface earlier, in a form someone could investigate.
What I did
Investigated recurring pipeline failures with SQL audits and root-cause analysis, validated large datasets, and found schema mismatches that were affecting downstream workflows.
What shipped
A small, inspectable validation flow (clean, detect, review) for an LG Brazil engagement, plus SOPs and runbooks for support and QA so recurring issues were handled the same way every time.
Result
Contributed to a 28% improvement in average incident-resolution time. Shared as an anonymised method, with no client data or proprietary code.