Karthik Kumar

Building Distributing Tracing infrastructure as a member of Observability Engineering.

• Currently developing a new system for trace data storage, modeled after data lake architectures. 
• Lead the design and implementation of a new system for trace collection. Developed a sidecar/collector to support higher volume, more reliable transport with lower overhead
• Developed long-requested aggregation and analytics features in the platform by deriving metrics from trace data, using Druid and Kafka
• Introduced a new form of sampling for trace data allowing non-ingress, mid-tier applications to make local decisions about sampling a request. Implemented this in platform libraries and rolled this out across the Netflix fleet (design)
• Scaled tracing infrastructure by 10x to handle increased load from new domains
• Migrated the core storage system for trace data from Elasticsearch to a managed time-series abstraction, which reduced cost by ~20% while improving search capabilities 

Built tools for robust root-cause analysis of latency and errors based on distributed traces, system metrics and logs from spans.

• Core contributor to an industry-first product that provided flexible analysis of large collections of traces (blog)
  • The feature allowed complex filtering and grouping of distributed trace data and identified system attributes that are correlated with high latency & errors
  • As part of this project, designed and migrated a core monolithic service into a distributed leader-worker model that horizontally scales to meet demand
• Developed a product to visualize system architecture as a nodes & edges diagram derived from distributed trace data (blog)
  
• Led a project to integrate with other open-source tracing libraries (Jaeger, Zipkin) to improve data ingestion flexibility and capability
• Proposed and prototyped several ideas that expanded analysis capabilities and were eventually integrated into the product

• Led a team of six engineers building features for Rally Engage. A consumer product that promotes healthy habits through rewards & incentives
  
  • Built a robust recommendation system to suggest relevant activities based on a user’s health goals and habits which led to increased engagement and participation 
  • Architected and implemented a foundational system for translating strings on-demand based on a user’s locale
  • Migrated a legacy, monolithic application into several scalable, independently deployable services. These microservices had lower latency, higher test coverage and better error handling (blog)
• Member of Eng. Tech Staff, serving as a technical leader on many large initiatives
  • Led a "Night's Watch" team of ten engineers that triaged and fixed critical performance bottlenecks and successfully scaled services to meet increased demand on Jan 1 

• Core contributor to a data integration platform for ingesting 100 billion utility data points/year from over 100+ unique clients (blog)
  • Migrated a legacy data import framework into a batched process, which reduced data import time from several hours to a few minutes 
  • Developed several data validation techniques, improving data quality and accuracy and tools to visualize ETL process and surface data quality issues to client