At Salesforce, I’m a core engineer on the Flink-as-a-Service platform and Cipher, a streaming ETL pipeline serving tenant teams through Kafka. I architected Cipher’s multi-tenant concurrency engine with custom Flink operators and redesigned checkpointing, reducing latency 8×.
I led Cipher’s upgrade from Flink 1.16 to 2.1 and Java 11 to 17 across stateful nodes. I also diagnosed RocksDB native-log bloat and off-heap memory pressure, helping reduce daily platform restarts from around 100 to single digits.
As Cost-to-Serve Lead for the Streaming Platform, I introduced autoscaling for non-peak workloads, right-sized job topologies, and drove migration to AWS Graviton, reducing annual spend by about $3M. I’ve also contributed to Apache Flink, including authoring FLIP-545 and resolving a native memory leak in the Flink Window Operator.
Before my platform work, I authored more than 10 peer-reviewed papers in applied ML and NLP at AAAI, ACL, EMNLP, and WWW. I presented at Flink Forward 2025 and received the Best Paper Award at AAAI AffCon 2020.
