Experience

Own the product direction and architecture of Supabase Pipelines. Defined the initial pricing model and led delivery through public alpha; the platform now moves terabytes of data per month.

Own Kubernetes infrastructure development, cost optimization, SLOs, on-call, and incident management.

Designed a crash-safe schema change propagation system on top of PostgreSQL logical replication, which has no native schema-change support.

Built a parallel table copy algorithm that partitions the initial table to speed up copy performance.

Built memory-aware dynamic batching that measures process memory, sizes batches, and applies backpressure to prevent OOMs and keep replication fast under memory pressure.

Hired engineers and grew the team from one to six, establishing the technical interview process and interviewing tens of candidates.

Rewrote the Supabase ETL prototype from scratch into a production-grade PostgreSQL logical replication engine in Rust. Owned early design, architecture, deployment, and infrastructure.

Increased BigQuery ingestion throughput 10x by redesigning the write path around HTTP/2 multiplexed bidirectional streaming with high concurrency.

Built unit, integration, and simulation testing infrastructure, increasing the team’s iteration speed and confidence in shipping new features.

Implemented the observability stack, enabling the team to detect and triage errors within minutes of their occurrence.

Designed an extensible open-source engine used by Supabase and other companies. Wrote extensive documentation to support community adoption.

Co-designed Relay’s partitioned SQLite disk buffer for an ingestion path processing half a million events per second, partitioning work across independent stores.

Built a bounded async task pool in Rust and migrated Redis consumers to asynchronous connection pooling, fixing connection starvation from holding clients across await points.

Built Sentry’s metrics query API and bulk executor, running independent queries concurrently while preserving dependencies and result alignment.

Built a durable alert migration system that moved every alert onto new infrastructure with zero downtime.

Designed and implemented dynamic sampling v2, using retention heuristics to decide which data to sample, reducing costs while preserving data quality.

Implemented JavaScript stack trace symbolication using debug IDs to resolve source maps, along with a new indexing system for those maps.

Graduated with 110/110 cum laude.

Thesis on distributed optimization with Apache Spark and Kubernetes, which led to two peer-reviewed papers presented at FiCloud 2022.