Distributed Computing @ Venmo

As part of the Venmo Risk team, I help build highly scalable systems to make efficient, realtime risk based decisions.

Improved customer experience by drastically slashing P99 latencies by 50%.

Revamped critical infrastructure by moving away from complex systems to highly-concurrent Java based GRPC APIs, reducing error rate by 10% and improving data-availability to allow our machine learning models make more informed risk decisions.

Cutdown mean-time-to-production to 2 hours from a couple of days by moving to independently delivered services. Used the opportunity to revamp the service by improving latencies across the board by 20%, improving CI/CD infra to fully automated to reduce deployment triggered incidents.

Found network related latencies within our stack that were adding 100% overhead to our p95 & p99 calls and moved off to closer and more reliable network edges to reduce latency by 60%.

Added observability to all our infrastructure and found long living bugs that hurt quality of our risk decisions.

Rearchitected async architecture in risk, utilizing managed AWS services to remove barriers to scale and reduce on-call and operational pain.

Mentored junior devs in the company with engineering best practices, both through example and directly.

Encouraged innovation and knowledge sharing within team through regularly scheduled tech talks.

Started formal process to document what we own through both code & design docs.

Lead by example by adding o11y to every significant code branches we touched.

Created highly secure apps that abstracted away secrets management, which later was adopted by the company as the preferred way to manage & deliver app secrets

When I joined metromile, we had a technology that worked but was slowly buckling under load. As we scaled, we ran into scalability and infrastructural issues that not only put our tech at risk but also slowed down delivery of features to our customers. As a part of backend team I tackled these challenges by actively revamping our backend infrastructure & reducing tech debt even when sometimes it meant moving beyond our squads’ charter.

I lead break down of business critical telemetry pipeline from our monolithic application into multiple micro-services - from integration with device provider, to storing normalized data per individual device for each customer. New services were designed to cater for legacy data as well, for seamless migration. These new services then allowed faster delivery, with each service being independently scalable. Added proactive monitoring and instrumentation to have full view into performance of these critical services. All these services were critical for revenue generation so were designed with failure tolerance.

Lead a team to allow auto-insurance quote aggregators to seamlessly integrate with our internal rating engine for quote comparisons while keeping our internal data models obscure. Built new services to allow for external facing integration (both ingress & egress) with swagger powered documentation.

Added multiple micro-services to our stack to separate out critical codepaths from non-critical ones.

Added shared indexes between micro-services to optimize business object sharing between independent services.

Primary technologies used: Java with AWS hosting(springboot, SQS, S3, Dynamodb, ElasticSearch, Kinesis, Postgres, MySQL, Docker, ECS, Fargate, redis, graphql)