At Toppr, I owned an Airflow ETL platform processing around 15 million learner events a day. I traced bad records, delays and schema drift to their causes, then shipped validation and retry fixes to prevent them recurring.
Across Western Union, I owned daily ETL for around 20 million financial transaction records, writing Python and Bash validation and reconciliation routines for every load. At FutureLearn, I ran an online-learning data platform and made pipeline failures and data freshness visible with Prometheus, Grafana and ELK.
Most recently at Turing, I owned the operational side of LLM systems, deploying containerized model-serving components through CI/CD to Kubernetes and building observability for latency, throughput, errors and GPU use. I’m now working on LLM infrastructure, where training-data quality is the same problem in different clothes.
