Project notes
I owned the pipeline for a third-party dataset containing billions of dealership repair-order records. I used Gemini with web-search grounding to standardize part descriptions before ACES mapping. Inputs included vehicle details, part numbers, and mechanic descriptions.
I combined repair-order data with expert-curated Probable Causes in BigQuery and applied expert-defined related-vehicle rules to extend coverage. A teammate owned the Probable Causes pipeline.
I deployed Cloud Run Jobs and created Airflow DAGs to orchestrate monthly publishing to AlloyDB. My work also included contributions to prompting and structured-output validation, with repeated subject-matter expert reviews.
Known Fixes launched as a core Diagnostic Intelligence feature. Diagnostic Intelligence reached 1,700+ subscriptions in its first three months.