Led a project to track and manage demand within IGML. This included implementing distributed tracing for client cost attribution, aggregating metrics across shared services, and creating a standard system for regression detection and admission control. Also delivered team wide keynote and workshops on managing resources solving future regressions, and best practices

Led efforts to map the data lineage of IGML’s ecosystem into a knowledge graph then delivered a stakeholder keynote on how to leverage it for better dependency management, privacy control, and cost benefit analysis

Led a project to restrict access and enforce deletion of personally identifiable information

Rearchitected IGML client logging and realtime processing services into a more efficient layer that reduced host footprint by 25%

Cut realtime feature storage footprint by 20% by deleting features with low model significance

Optimized feature storage serialization and batched key records based on access pattern to reduce footprint by 30% and cache latency by 15%

Migrated ML features to new framework which allowed the deprecation of a legacy feature framework- saving 75% of database write QPS, 10% of fleet size, and increased engagement by 0.7%

Built and scaled a custom, in-house 3D render job pipeline

Saved customer success 20 hours/week by creating an order management system with automated product refunding

Improved query times by 50% and unified data access patterns by creating a data access layer for products and 3D render jobs

Refactored and removed 20,000 lines of code within the first 3 months

Increased unit test coverage from 58% to 80% by standardizing good coding practices and review

Improved site reliability after benchmarking and refactoring expensive APIs and queries

Created a release tool to automate deploys and release note creation that decreased deploy time by 5 hours/week

Created an asynchronous job to parse and convert 3D asset files into V-Ray renderable files

Provided real-time support for critical bugs using Bugsnag, Papertrail, Fullstory, and CloudWatch

Increased workforce throughput by using Celery to asynchronously batch Amazon Human Intelligence Tasks (HITs) based on perceived effort, task type, and completion time

Decreased job latency by building near real-time repricing of undesirable Amazon Human Intelligence Tasks (HITs)

Designed and created a new regex model to support forms with pages that are optional, repeating, and out of order

Increased system reliability by helping migrate a part of job pipeline into a separate microservice

Decreased system downtime through throttling and splitting of large jobs that consumed excessive resources

Improved customer turnaround time using job prioritization based on service level agreement (SLA)

Ensured system scalability via benchmarking, optimizing Postgresql queries, and segregating Celery infrastructure

Provided real time support during system outages using Airbrake, Splunk, NewRelic, and custom monitoring integrations

Mentored two summer full-stack interns and helped them deliver core improvements on form setup and configuration

Built a JSON configuration compiler to chain Hadoop MapReduce jobs and perform cohort analysis on large datasets

Implemented client retention rate and "layer cake" plot visualizations using d3

Created a custom LRU Redis cache with locking for parallel operations and manual garbage collection