Muthian George Email & Phone Number

Mountain View, California

Definition of SQL-like stream processing scripting language, developing compiler for reactive stream data processing in IoT edge devices:SQL-like reactive stream processing language definitionCompiler design and development – as an interpreter as well as compiled program via C++ code generation.Building hundreds of high performance statistical and machine learning functions in IoT edge devices for easy use through the stream processing scripting language.Building a complete matrix and measures (all seven dimensions) libraries for high performance stream data processing.Integration of string pattern matching, string processing, math libraries and a number of utilities into the functional script language.Built parser for PMML for generating executable code so that machine learning models are developed with large set of data in the cloud and deployed for prediction in IoT controllers. ML models were built for regression, general linear regression, k-means, nearest neighbor, SVM and tree models.XGBoost library integration for machine learning models.Stream data join processing for n-streams based on data field such as timestamp with fudge value.

Sunnyvale

Building and upgrading client/server software products of small and medium business. Built libraries to isolate code from yahoo main branch for company spin off and for performance improvements.

Redwood City

Development of user defined functions framework in C++2011 for ParAccel, vector and ingres databases for data processing using SQL.Classification of analytical functions for integration into the databases. Aggregate functions are processed using GROUP BY and PARTITION BY clauses in a distributed environment. Table valued functions are used for accessing external data from HDFS or other databases for federated query processing and data loading.Development of vector processing framework to exploit SIMD computations for high performance data processing.Aggregates are supported as single phase or multiphase functions. Multiphase functions are used in MPP clusters where each computing node accumulates intermediate results and passes them to the leader node for producing the final result.UDF framework runs inside the SQL engine or outside in a sandbox server that supports many concurrent queries.Built an application hosting server using TCP/IP.

Palo Alto

Leading analytics team in the SeaQuest/NeoView MPP Database Group.Built a user defined function (UDF) framework for the integration, development, management and deployment of externally built UDF libraries as shared objects for processing in databases, HADOOP, HIVE and any other analytical data processing system.Defined SQL syntax and processing semantics for integration of UDF framework within SQL database so that SQL is elevated for analytical processing.PatentsUser Defined Function Classification in Analytical Data processing Systems. (USPC Patent Application Number: 20120191642 Publication Date: 2012-07-26. By Muthian George.Parameter Expressions for Modeling User Defined Function Execution in Analytical Data Processing Systems. (USPC Patent Application Number: 20120191690 Publication Date: 2012-07-26. By Muthian George.Aggregate Function Partitions for Distributed Processing. (USPC Patent Application Number: 20120191699 Publication Date: 2012-07-26. Muthian George, Song Wang & Ahmed K. Ezzat.Output Field Mapping of User Defined Functions in Databases. (USPC Patent Application Number: 20120191732 Publication Date: 2012-07-26. By Muthian George.User Defined Functions for Data Loading. (USPC Patent Application Number: 20120239612 Publication Date: 2012-09-20. By Muthian George & Song Wang.

Fremont, California

Built a reactive stream processing real-time system for high frequency trading.Built a Complex Event Processing (CEP) server engine for real-time stock market analytics. Data are structured as vectors for high performance computing in SIMD hardware architectures. CEP engine was modeled as events (data arrival), analytics processing, conditional evaluation and action generation. Algorithms were composed and set in the CEP engine for automating system monitoring for buy and sell signals.Built the system for pipelined processing in multicore machines in core-bound threads in a thread-pool to exploit CPU cache-hits..Built around 200 time-series analytical functions and candle-stick patterns used mostly in the technical analysis of equities.Built systems to back test system for analysing millions of algorithm combinations to find highly performing algorithms in trading.Built multiple monitor visual tools (usually 4 monitors and 16 charts and each chart with multiple indicators) for online monitoring of the market. Also built a number of stock filtering algorithms and compare and contrast results.Built a cluster with master/slaves for hosting market applications using TCP/IP and multicast protocols.

San Mateo

Business Intelligence Suite: Envisaged and built the product suite:Built a federated business intelligence data processing engine. Aggregated enterprise data from multiple sources, ETL processed dynamically on demand. Built ROLAP, MOLAP and managed query models on top of cleansed and transformed data.Built physical metadata for source schema, and logical metadata for ETL and multidimensional dimension application metadata. Analytic metadata were built with functions on measure fields with dynamic dimensions.Built the complete SQL-92 standard query engine consisting of multiple table operators for joins (inner, and left, right and full outer), union, union all, union join, except, except all, intersect and intersect all.Built a multithreaded concurrent server to host the SQL query engine where many concurrent queries could be run. The query parser was objectified after yacc created the parser for concurrent query processing.With Oracle, the product generates a billion dollar annually.