Jan 2022 – Jun 2022 · 5 mos · Sunnyvale, California, United States

• I ran engineering for "Streams and Lakes" - the data integration, streaming, and open source analytics services on Google Cloud. These include Dataproc, Dataflow, PubSub, Dataplex, DPMS, Data Catalog, Composer, Data Fusion, and more. Customers use these "first mile" services to move, prepare, organize, and analyze their data in data lakes and data warehouses.

Jan 2020 – Jan 2022 · 2 yrs

• I ran the Amazon OpenSearch Service (successor to Amazon Elasticsearch Service), the OpenSearch project, Amazon CloudSearch Service, and AWS Lake Formation. I was responsible for go-to-market, product, engineering, and operations. I helped set strategy and oversaw the fork of OpenSearch from Elasticsearch.
• I was part of analytics at AWS. We were re-architecting big data systems for the cloud, where resources are plentiful, data is abundant, and everything is a service. I had a chance to put my crazy ideas into practice and deliver them to thousands of customers.

Jan 2015 – Oct 2020 · 5 yrs 9 mos

• I launched and ran engineering, operations, and go to market for two fast-growing cloud services: AWS Lake Formation and AWS Glue. We are a multi-disciplinary team of distributed system and database architects, front-end and UX specialists, and ML experts.
• AWS Lake Formation is a new service that makes it easy to setup, secure, and manage Data Lakes. Data Lakes -- the evolution of enterprise data warehouses -- are curated repositories of structured and unstructured data that allow self-serve analytics for modern use-cases: IoT, ML model training, data science, and more. AWS Lake Formation leverages cutting-edge ML techniques for data ingestion and cleaning, and simplifies data security and governance. We envision it as the locus of control for all data in an enterprise.
• AWS Glue is a serverless data integration service which powers AWS Lake Formation. AWS Glue offers a centralized metadata service -- Data Catalog -- with crawlers that automatically extract and index metadata to enable data discovery. It also provides an ETL (extract-transform-and-load) engine that automates much of the undifferentiated heavy-lifting for moving and transforming data sets. At its core, it uses the Schema-lift technology from Amiato that makes it easy to handle modern semi-structured data sets.

May 2014 – Sep 2015 · 1 yr 4 mos

• I led the development and public launch of two headlining features in Amazon Redshift: interleaved sort keys (Z-indexing) and user-defined-functions (UDFs). Interleaved sort keys are an alternative to indexing and projections for columnar databases that allow fast search on tables across multiple dimensions. UDFs allow users to customize Redshift analyses in Python for modern big-data use-cases.

Jan 2018 – Jun 2018 · 5 mos · Berkeley, CA

• In the Spring 2018 semester, I taught the Introduction to Database Systems course, an upper-division course for juniors and seniors in the EECS department. The course had 450+ students enrolled and a staff of 10 TAs.

Sep 2011 – May 2014 · 2 yrs 8 mos · Palo Alto, CA

• Amiato was a fully managed, real-time ETL cloud service. It bridged the gap between unstructured data and the world of structured business intelligence (BI) tools. Schema-lift was the technology powering Amiato. Schema-lift automatically infers the structure of semi-structured data (e.g. JSON logs), transforms it into tables, and loads data warehouses.
• As CEO and co-founder, my responsibilities spanned: fund-raising, recruiting, sales, go-to-market, and managing customer and partner relationships. I led the M&A process for our successful acquisition by Amazon. I raised our funding which included notable investors: Andreessen-Horowitz, Ignition, and YC. I also helped build early prototypes and design Schema-lift.

Sep 2004 – Sep 2011 · 7 yrs · Palo Alto, CA

• My work at HP spanned large-scale data management, distributed computing and energy-efficiency. Below I list my most significant projects in reverse chronological order.
• Armonia: Principal investigator for Armonia project -- a scalable, distributed, main-memory data management platform that offers strongly consistent low-latency operations and complex on-the-fly analytics. Applications include financial trading and social networking.
• HP-KVS: Built a highly available, low-cost key-value service for the cloud. HP-KVS is an eventually consistent, erasure-coded, large object store that spans multiple geographies.
• Sinfonia: A highly scalable, distributed transactional store for building data-center infrastructure applications. Built a large-scale distributed B-tree, clustered file-system, and group communication using Sinfonia. Basis for the Armonia project. Won best paper, SOSP 2007.
• Energy-efficient systems: Characterized and optimized the energy use of computer systems as a whole, from storage to memory to compute. Inventor and maintainer of the JouleSort bechmark, the first holistic energy efficiency benchmark, which has inspired efficient server designs and influenced other benchmarks. Investigated energy efficiency of DB workloads.
• Other work: Designed software and hardware for non-volatile RAM technologies like NAND Flash and Memristors. Developed methods for long-term preservation of digital information.

Jan 1999 – Oct 1999 · 9 mos · Almaden

• Investigated alternative strategies for implementing collection types in IBM DB2/UDB. Designed language extensions for querying collection types. Gained experience with administration and software development in DB2/UDB.

Jan 1997 – Jan 2004 · 7 yrs · Berkeley, CA

• Thesis: “Flux: A Mechanism for Building Highly-Available, Fault-Tolerant, Scalable Dataflows”: In the TelegraphCQ system, my dissertation focused on making parallel CQ dataflows – computations that analyze high-throughput streaming data in real time – highly available, fault-tolerant, and automatically load-balancing.
• Continuously Adaptive Continuous Queries (CACQ): Developed an adaptive query processing system that executes numerous long-running queries simultaneously over streaming data.
• AMDB: A debugger and profiler for search indexes on non-traditional data types like audio and images. Designed UI for navigating high-fanout search trees. (Released open-source).

Jun 1996 – Jan 1997 · 7 mos · Murray Hill, NJ

• MEng Thesis (jointly done at MIT): "ReferralWeb: A Resource Location System Guided by Personal Relations." The first system to automatically discover and extract social networks by mining publicly available data on the web. ReferralWeb also automatically finds experts on user-specified topics and recommends paths in the extracted social graph to connect users with those experts.

Jun 1995 – Aug 1995 · 2 mos · Murray Hill, NJ

• Built a prototype of a content-based image search and retrieval system.

Jun 1994 – Aug 1994 · 2 mos · Holmdel, NJ

• Built tools to integrate simulated navigation systems with context-relevant websites.