In collaboration with my Search team colleagues, we are responsible for the design, implementation, operation, maintenance, and evolution of all products search functionality in a large e-commerce site, which sells and delivers-to-your-door-step products for ~200 retailers in the food and drug, grocery, retail pharmacy, department store, and other everyday consumer goods, totaling millions of products.Our search system is composed of:- New and updated products ingestion pipeline from catalog to search engine;- Machine Learning (Learning-to-Rank) features generation and ingestion-to-search engine pipeline;- Search (including auto-complete) micro-services and their APIs.- Miscellaneous search-supporting data stores.Technologies: Go (APIs), Elasticsearch (Search engine), Kafka (Data pipelines), Redis (Cache), Bigtable (Big data store)

I am a search consultant in the design, development, deployment, optimization, and relevancy tuning of search applications built with the open source search engines such as [Solr](http://lucene.apache.org/solr/) and [ElasticSearch] (https://www.elastic.co/).Below are three examples of significant client engagements I was intimately involved in:1. [1 year] Complete revamping of the ranking logic and implementation of a public-facing regional online digitized “Yellow Pages”. Highlights:- Custom scoring & ranking based on a variety of "relevancy signals" based on matching levels (“high-precision” or “fuzzy” mm-based business name match, exact category match, etc.), business “spend” by Yellow Page client, geo-location, as well as classic Information Retrieval text-based BM25-scoring.- Relevancy-ranked faceting.- Dynamic function-based grouping.2. [3 years] Design and development of an open source-based next-generation search platform in a large multinational information retrieval company.- Custom Solr query parser (Boolean, proximity, clause frequency, as-is, etc.);- Custom Lucene analyzers for term-confidence level association, and “as is” matching semantic;- Custom Lucene queries for confidence level-based matching using Lucene payloads;- Methodologies and supporting tooling for Recall & Ranking testing and tuning, as well as performance analysis and tuning;- Custom LTR features extraction plugin development and performance tuning;- Very large SolrCloud (2B documents) architecture, AWS-based pricing options, and perf. tuning (Topology, sharding, cloud computing resources);- Overall solution “hardening” and preparedness for releases to production;- Top 10 committers over the past 3 years within the Search Platform git repo.3. [6 months] Design and development of a "multi-ranking" system in an e-commerce application to allow for competing ranking criteria such as popularity, recency, and various forms of diversity across search results.

I continued to support my client's Data Warehouse system, develop new data reports, and increase the level of build automation for the development and QA environments using Maven to manage a host of database/schema profiles, Ant to execute the SQL DDL and DML build scripts for a variety of "targets", and Jenkins to schedule and execute the Ant targets.

SRA and CSGov merged in November 2015 to form CSRA.In the summer and fall 2016, I developed and deployed to production a Groovy-based code generator that generates the PL/SQL source code for Oracle APEX-based data grid Web pages. The generator utilizes metadata in the client's data warehouse data dictionary to specify all attributes in the data grid: Fact table (i.e.. the SQL select statement), columns orders and headers, sort order, etc.In the Spring 2016, I developed a Jenkins-based automated build of the client's data warehouse system for the development, QA, and staging environments. The automated build, to be run at any desired frequency, re-builds the data warehouse's various database schemas DDL, metadata, and data dictionary from a Git-based source control repository. The build is executed via a series of coordinated Jenkins jobs.In march 2016, we released version 1.1 of the client's QA Checks Engine (a.k.a., FDSS, started in June 2015 at SRA) that processes industry-submitted reports across a variety of regulations and programs on fuels. The FDSS system applies a number of client-defined QA Checks on the submitted reports, persists them in an operational (normalized) database, as well as moves the processed reports to the client's data warehouse to support reporting and data analytics.The system's "data pipeline" goes as follows: [Industry Submitted/Staged (Excel-based) Reports] --> [FDSS (QA Checks)] --> [Operational Database] --> [Data Warehouse] --> [Reporting & Analytics User Interface].One interesting characteristic of the system is the ability to rebuild from scratch, or "reproduce" (Using Data Science best-practice), the data warehouse after making changes in the QA checks (i.e., business rules).Technologies: Grails 2.5 (QA Checks Engine application), Tomcat 7 (Application container), Oracle 11g (Database & Data warehouse), APEX 4 (Data warehouse UI), Jenkins (Continuous Integration).

Building on the success of the migration of the legacy FFARS data in the first part of 2015 (See previous project at SRA), I led the design and development of a "QA Checks Engine" to quality control the client's industry-submitted Excel-based reports in a legacy (unstructured) database, and move the processed reports to the client's data warehouse for viewing, mining, and analyses. The version 1.0 of the resulting system, coined FDSS for Fuels Data Submission System, was put in production in December 2015.Technologies: Grails 2.5 (QA Checks Engine application), Tomcat 7 (Application container), Oracle 11g (Database & Data warehouse), APEX 4 (Data warehouse UI).

Starting in January 2014, I led the complete re-design and implementation of EPA's Fuels and Fuel Additives Registration System (FFARS) relational (Oracle) database and Web application (Grails). FFARS is an internal EPA application for EPA staff to perform manual data entry for the registration of all fuels and fuel additives used in motor vehicles in the US. Leading a team of 5, we reached a features-complete milestone a year later in December 2014, then completed the migration of the legacy FFARS data to the new re-designed database in May 2015, date at which the system went live. The salient point of our legacy data migration approach is that the migration software was written as part of the new Grails application (as a separate service), and utilized the Data Staging, Quality Assurance Checks, and Registration services that were implemented for the normal day-to-day data entry processes. Our legacy data migration approach was particularly successful in terms of leveraging the application's existing services, as well as utilizing Groovy and Grails' GORM as a potent RAD for general-purpose data extraction, transformation, and loading (ETL).

SRA acquired PQA in May 2010. Within SRA, I continued to lead the development of the EMTS system at a pace of 3 to 4 releases to production a year. In mid-2012, per our client's request, I switched my focus to data mining and reporting.I led the development of a solution for data warehousing and analytics across a variety of programs and operational databases within our client's Office of Transport and Air Quality (OTAQ) division. The solution was implemented in Oracle's Application Express (APEX) rapid web application development as well as Oracle's Business Intelligence (OBI) frameworks, with an ETL layer implemented in PL/SQL.

I lead the architecture and software design, development, and production support of EPA OTAQ's Moderated Trading System (EMTS). EMTS supports EPA's Renewable Fuel Standard regulation (http://epa.gov/otaq/renewablefuels/index.htm).EMTS was implemented in Java using open-source frameworks such as Maven (Build and deploy), Spring (Dependency Injection, and Aspect Oriented Programming), Hibernate (Object/Relational Mapping), Struts (Web tier), and ActiveMQ (Enterprise Messaging), and using an Oracle 11g database.In production since July 1st, 2010, EMTS has been a "success story" for EPA: The system has vastly simplified and automated a number of EPA tasks and reports; it has been reliable with only a few production issues which were resolved promptly; and the system has been well received by the industry. In addition, EMTS has performed well with an ever increasing traffic: The system currently processes an average of 25+K "EMTS transactions" per day, two-third of which are submitted as XML files, and the remaining one-third being submitted from the EMTS Web application.Note: An "EMTS transaction" is an order placed by an industry user to perform a certain operation on "RINs". RINs are the renewable fuel credits that EMTS manages.

- Led the design and development of Vertical’s next generation Wave™ IP Telephony “platform,” which includes all hardware, cards Linux firmware, and Windows and Linux telephony drivers.- Managed an engineering staff spanning three locations and including eleven hardware engineers and their manager (encompassing the Electrical, Components, CAD, Mechanical, Test, and Doc Control engineering functions), three embedded Linux firmware engineers, and three Windows drivers engineers.- Released the Wave IP 2500™ and 500™ platforms (http://www.vertical.com/products-wave-ip.html) on schedule and on budget.- Responsible for the sustaining of the Comdial's telephony products (FX II and MP5000 IP PBXs, EP100 digital phones, iPrimo and EP300 IP phones, EP200 Windows-based SIP soft phone, and Interchange unified messaging products.)

• Led Comdial’s research and development efforts, having corporate-wide responsibilities for the architecture, design, development, QA, delivery, and sustaining of all Comdial telephony and messaging products: 4 proprietary business telephone switches (PBXs), 3 Voicemail products, and a variety of digital and IP telephones.)• Established Comdial’s product architecture vision and road map to transition Comdial’s products from digital telephony and messaging systems to a next generation Internet Protocol (IP) Telephony-based architecture.• Released several significant products for Comdial and its nationwide family of resellers:- 2004: MP5000™ IP PBX, EP300™ SIP phone, and EP200™ Windows-based “soft” phone (http://www.vertical.com/products-mp5000.html.)- 2002: FX II™ networking support up to 14 nodes.- 2001: iNTerchange™ unified messaging.

Led the research, design, and development of Comdial's next generation hybrid digital switch/VoIP business class telephone systems, leading to the foundation for Comdial's MP5000(tm) hybrid TDM/IP telephony products. Developed a SIP stack and Proxy prototypes in Java.

Led all phases in the vision, specification, architecture, design, implementation (C/C++/COM), QA, and deployment of Comdial’s award winning iNTerchange™ Unified Messaging product. I personally designed and implemented a C++/COM MAPI/Outlook-based agent service that synchronizes voice, fax, and email messages between an email store and a voice mail store. The underlying two-way synchronization algorithm that I developed was unique in the Unified Messaging industry and was award US Patent #6792085.

Acted as Project Manager for Comdial's voice mail products(ExecuMail, Eloquence.) Led the co-development between Comdial and Rhetorex (www.rhetorex.com) of a digitally integrated PC voice processing card (a.k.a., IVPC.) Led the R&D effort to establish Comdial's voice processing road map in terms of overall strategy, features, architectures, and design. Assessed the technologies and products developed by the Key Voice Technology, Inc. (KVT), a leading voicemail company at the time, which led to the acquisition of KVT by Comdial in 1996, which has been touted as Comdial's best acquisition.

Designed and implemented a Windows-based user interface for a pen-operated and portable computer application. The system was delivered to SOFRES (France's leading public opinion agency) in January 1994 and deployed to 300 itinerant interviewers.

Lead software engineer in the design and development of an automated telephone- and speech recognition-based public opinions polling system (a.k.a., "CARUSO"), which was built for France's leading polling agency SOFRES (http://www.tns-sofres.com/).