Google Analytics, Attribution, F1Search Ads 360 / Google Analytics 4 (GA4) Integration. Our customers needed the GA4 conversions attributed to the SA360 campaigns for reporting and bidding. Led the backend integration. $15+ billion in linked revenue. https://support.google.com/analytics/answer/11085214Google Optimize / GA4 Integration. Optimize customers needed their A/B and multivariate tests to work with GA4 traffic. Led the backend integration. https://support.google.com/analytics/answer/11378318Attribution for Google Ads. Ads campaigns needed more sophisticated and realistic attribution models. Worked in a team that provided various attribution models to Google Ads, including a machine-learning generated model based on each customer's specific traffic. Impacted $100+ billion in revenue. https://support.google.com/google-ads/answer/6394265Attribution for GA4. Google Analytics traffic needed to be attributed to customers' Google Ads search campaigns, e.g. how many conversions or sessions did this campaign generate. Built a high throughput service (peak 1.8 million QPS) capable of attributing all Google Analytics traffic.Google Analytics. Our customers needed higher cardinality tables than what Google Analytics was providing. Built 20X cardinality of 1MM. https://support.google.com/analytics/answer/1009671F1, distributed query engine.- Dynamic load balancing. Built a dynamic distributed scheduler that reduces the long tail and improves execution time (work stealing in a distributed environment).- Query redirection. Built a query redirection mechanism that selects the closest data center to the referenced data and sends the query there.- Added first-class support for columnar storage in F1 query engine- Built gaping functionalities in GoogleSQL and helped migrate Ads to GoogleSQL .Google readability in C++ and SQL. Proficient in Python.Awards/bonuses: 1 engineering excellence award, 8 spot bonuses, 20 peer bonuses.

User Analytics for Social and Mobil Web. Implemented an HBase/Hadoop data store in Java that saves state in a distributed cluster for batch/stream data processing. Implemented REST API and server for this data store.

Conducted data and performance analysis for Smart Meter Analytics.http://www.youtube.com/watch?v=hgTGwNNfPpIConducted performance analysis of SAP's core business in-memory database engine on 32-, 64-, and 128-core machines on TPC-H and SAP benchmarks ( up to 8 TB of data) for query latency and throughput optimization. Implemented monitoring tools, scripts, and visualization for Linux x86-64 ccNUMA on various metrics, such as threads affinity and NUMA data placement.http://www.youtube.com/watch?v=kwu5fndwz9Ahttp://www.youtube.com/watch?v=uUqtUw-m7mQLed the design and implementation of a scalable C++ library of user-level threads for x86 manycore architecture on Linux using non-blocking synchronization techniques that outperforms Linux native POSIX threads on PARSEC benchmarks. Implemented threads, synchronization primitives (locks, condition variables, barriers, rwlocks), and non-blocking thread scheduling. Also evaluated performance against OpenMP and Intel TBB. Patented two parallel scheduling algorithms: 1) non-blocking work stealing for cooperative threads and 2) locality improving scheduling for ccNUMA multicores. Designed and implemented a scalable non-blocking multi-lock algorithm (tested on 64 cores)Conducted academic programs, and mentored interns.

Designed and implemented memory allocation techniques in C++ that improve an application's execution time an average of 7% and up to 22% over state-of-the-art allocators, such as Doug Lea's malloc and FreeBSD's malloc. Validated them on real world applications such as industry standard SPEC CPU2006 benchmarks and Fortran compiler. Integrated them into C++ Standard Template Library. Resulting method required only an application re-compilation. Designed and implemented a generic memory allocation library. 30K lines of C++.Researched, designed and implemented the C++ Standard Template Adaptive Parallel Library (STAPL), a parallel equivalent of C++ Standard Template Library (STL). Implemented parallel algorithms, containers and task scheduler in STAPL, for both shared (pThreads) and distributed (MPI) memory systems.Participated in the DOE ASCI Project - ”Efficient massively parallel implementation of modern deterministic transport calculations.” Implemented a communication/computation overlapping mechanism and performed locality improving optimizations.Teaching Assistant. “Programming in C”. Responsible for 3 sections of 20 students, supervised laboratories, graded projects, tests and other assignments.Fall 2003: Lectured ‘Compiler Design‘, senior level class. Lectured on lexical analysis, parsing techniques, context sensitive analysis, intermediate repre- sentations, procedure abstraction, heap management, code generation, instruction selection, code improvement techniques and optimizations.

Worked on hot spot empirical recognition in a C++ compiler (source-to-source translator, ROSE). Designed and implemented the outlining and generic instrumentation, both program transformations passes in the compiler.

“System Analysis and Design”. Responsible for 4 sections of 25 students. Lectured seminars, problem solving sessions, graded projects, tests and other assignments.