At Fornax Ventures LLP, I built the dbt-trino modelling layer for a Trino and Apache Iceberg lakehouse spanning ads, sales, returns, fulfilment, quick-commerce, traffic, customer retention and P&L. I designed partitioning and file ordering for ads data, then implemented incremental and micro-batch strategies that cut pipeline runtime by about 60%.
I engineered a PySpark attribution pipeline that moves GA4 events from BigQuery to Iceberg on S3, supporting six multi-touch models. I also traced a stalled backfill to a Prefect state watermark and an upstream build cap, restoring 27.5K attributed rows across the models and lookback windows.
For Frido, I set up an EL pipeline into BigQuery and refined dbt modelling and partitioned incremental loads, reducing scans from 115 GB to 5.5 GB. Earlier, at Dharma Diamonds, I developed Python and pandas batch pipelines and reconciliation checks across procurement, inventory, sales and accounting.
