About
Distributed systems engineer and cloud architect with a decade of experience designing fault-tolerant infrastructure for financial and data-intensive platforms. I combine a research background in consensus protocols with hands-on engineering to build systems that are both correct and performant. At Stripe, I led the migration of core payment processing to a multi-region architecture, reducing p99 latency by 34 % while maintaining six-nines durability. I am drawn to problems at the intersection of reliability, scalability, and operational simplicity.
Previously at MongoDB, I contributed to the distributed query engine and built tooling for cross-shard transactions. My work at Red Hat focused on performance isolation in containerized environments. I hold a Ph.D. in Computer Science from Stanford, where my dissertation introduced a novel approach to byzantine fault-tolerant state machine replication with sub-millisecond failover.
Experience
- Architected and led the multi-region active-active deployment of payment processing pipelines, reducing p99 latency from 210 ms to 138 ms and achieving 99.9999 % durability.
- Designed a distributed rate-limiting service using Redis Cluster and consistent hashing, handling 2.8 M requests/second with <5 ms overhead.
- Built an internal chaos engineering framework (Gremlin-compatible) that runs weekly fault injection experiments across 600+ microservices; discovered 23 latent failure modes in the first quarter.
- Mentored four engineers through the promotion process and established the team's incident response playbook, reducing mean time to recovery from 34 min to 11 min.
- Authored RFC for the migration from synchronous gRPC to asynchronous event-driven processing with Kafka, resulting in 40 % reduction in tail latencies during peak traffic.
- Core contributor to the distributed query planner for MongoDB 5.0, optimizing cross-shard aggregation pipelines — reduced network I/O by 28 % for deployments with >50 shards.
- Designed and implemented the transaction coordinator for multi-document ACID transactions across sharded clusters; the feature unlocked $12M+ in new enterprise contracts.
- Developed a benchmarking harness (written in Go) that simulates production workloads at 2 M ops/sec, used by the SRE team for capacity planning and regression detection.
- Led the migration of internal service mesh from Envoy to Linkerd, cutting control-plane latency by 60 % and reducing operational toil by 15 engineer-hours per week.
- Developed CPU and memory isolation controllers for the Linux kernel's cgroup v2 interface, directly impacting the performance predictability of containerized workloads in OpenShift.
- Built a telemetry agent (in Rust) that collected kernel-level metrics at 1 ms granularity with <0.5 % CPU overhead; adopted by the OpenShift cluster monitoring team.
- Contributed to the design of the Red Hat Enterprise Linux real-time kernel extension for financial services workloads, achieving a maximum jitter of 12 µs under load.
- Authored three kernel patches accepted into mainline Linux for improved memory accounting in cgroup namespaces.