Our technology will be the next big thing since Cost Based Optimization. We call it Workload Based Optimization. Through learning from users’ past workloads, we can make their future queries faster and cheaper.See http://getbluesky.io/.

Build runtime components for TensorFlow and Jax, supporting a range of accelerator devices (TPU, GPU, CPU, etc), across training and serving.

Build a new ML (TensorFlow) programming model in Swift for TensorFlow, combining ease-of-use with high performance: https://www.tensorflow.org/community/swift.See https://llvm.org/devmtg/2018-10/talk-abstracts.html#talk15 for a key underlying technology piece.

Lead a team to build and launch new unified storage and query serving infra for both (row-based) seek and (column-based) scan workloads.

- Build Hadapt analytic platform- Incubate analytic use cases and practice data magic

Close critical business deals; connect business with technology; incubate new analytic use cases. - Customer-facing eng work to solve business problems with Vertica, and close large deals (e.g. Zynga)- Identify new product opportunities through learning about user painpoints - Deliver tech talks and demos, to increase Vertica's market visiblity/influence, and acquire customers and eng talents (e.g. Hadoop world 2011 talk)

June 2010 - May 2011: Engineering TLM- Built and led a team of engineers to create and enhance the Vertica client software (drivers and BI tools connectors) for a smooth integration of Vertica into the Big Data eco system.- See http://bit.ly/v41driversJune 2009 - May 2010: Founder and TL of Vertica Database Designer- Lead the design, implementation and launch of a new component for automated physical performance tuning (US patent)- See http://www.vertica.com/company/news_press/31-vertica_extends_lead_in_analytic_data_warehouseJune 2008 - May 2010: Software EngineerLed the design and development of the following key components and features in the Vertica Analytics Platform:1. In-database analytics, including time series, sessionization, pattern matching and event series join (US patent). See the following blog posts:- http://www.vertica.com/blog/17-reading_between_the_lines_with_vertica_4_0- http://www.vertica.com/blog/12-gap_filling_and_interpolation_gfi- http://www.vertica.com/blog/11-more_time_series_analytics_event-based_window- http://www.vertica.com/blog/10-sessionize_with_style-part_1- http://www.vertica.com/blog/38-sessionize_with_style_part_22. SQL '99 analytic functions3. Various optimizer features, such as outer join rewrite, distinct optimization, group-by pipeline optimization with equality predicate.

Continued to work on the Microsoft CEDR event processing project, now known as the "SQL Server StreamInsight" technology (http://msdn.microsoft.com/en-us/sqlserver/ee476990.aspx). The work resulted in two more US patents TRANSLATION OF STREAMING QUERIES INTO SQL QUERIES and STREAMING OPERATOR PLACEMENT FOR DISTRIBUTED STREAM PROCESSING.

Co-founded the Microsoft CEDR event processing project, now known as the "SQL Server StreamInsight" technology (http://msdn.microsoft.com/en-us/sqlserver/ee476990.aspx). The work resulted in a CIDR '07 publication "Consistent Streaming Through Time: A Vision for Event Stream Processing", as well as two US patents CONSISTENCY SENSITIVE STREAMING OPERATORS and TEMPORAL EVENT STREAM MODEL.

Worked on the ImmortalDB transaction time database project.http://research.microsoft.com/db/immortaldb/Implemented the Time-Split B tree indexing technology into the SQL Server codebase, and conducted experiments to validate its efficacy. This work resulted in a VLDB '08 publication "Transaction time indexing with version compression".