Prasanth Kumar

I’ve spent 7+ years designing distributed data platforms on GCP and Databricks for teams at UPS, Ford, Mondelez, Loblaws, and Micron. I focus on the trade-offs that actually matter—streaming vs. batch, cost vs. latency, and correctness vs. throughput—especially for entity resolution, deduplication, and distributed skew handling at TB scale.

At UPS, I architected multi-source marketing ingestion into BigQuery, unifying Google Ads and Meta Ads behind a single incremental framework that handles quotas, auth, backfill limits, and schema drift. I also built a chain-of-custody dispute resolution system that correlates customer disputes with delivery scan events and proof-of-delivery imagery, using Databricks PySpark/Spark SQL plus Python services on GCP for idempotent re-runs and full auditability.

Before that, I led migrations into BigQuery for Mondelez, collapsing data-availability latency from hours to minutes using a hybrid of parallel batch ingestion and Pub/Sub-driven streaming. I’ve also migrated large Teradata and on-prem Spark workloads into BigQuery/Dataproc, reworking joins, window functions, and orchestrations in Cloud Composer to cut processing time, runtime, and cost while keeping pipelines safely re-runnable through replays and schema drift.