• Fixed several high-profile bugs in the Klout score MapReduce pipeline that incorrectly included
opted-out users in the score computations and public website.
• Added alerting that triggered when the scoring pipeline completed, so that adherence to SLAs
could be determined accurately.
• Developed ways of making the backend more efficient and realtime using the Storm framework
and Redis.
• Implemented a Storm bolt for persisting data continuously to HDFS.
• Wrote a custom Hadoop InputFormat to allow MapReduce jobs to consume data from Kafka
queues.