Dec 2018 – Oct 2019 · 10 mos · Bengaluru, Karnataka, India

• Designed and implemented generic Pyspark based package to seamlessly query S3 avro data over spark clusters seamlessly (without requiring any Spark programming ) like that is done over managed/server-less data warehouses/ lakes like Google BigQuery.
• Comprehensive study of how AI-enabled data centre management/network monitoring products are and should be including definition of AI/ML problems therein with the goal of incorporating such features into Cumulus’s netQ product, a network monitoring product providing real-time visibility and troubleshooting of modern data centre networks.
• Simple ARIMA based time-series trend prediction of Rx and Tx bytes of network switches- supervisory (not hands on)
• Anomaly detection for data quality over switch data

Sep 2017 – Oct 2018 · 1 yr 1 mo · Bangalore

• Designed/Implemented/Deployed a prediction model (avg. accuracy 93%) for generating large scale customer affinities to user attributes (price, brand) for e-commerce personalization with a repeat benefit data architecture for other data science problems. The architecture within which the solution was cast (BigQuery pipeline with custom BQ tool triggered by crons) also along the way builds hierarchically aggregated features over clickstream that can be instantly used for various data science problems, most importantly to avail labelled data that supervised learning algorithms need. Significant resources get saved as prebuilt cleaned datasets with diverse selectable features organised in a kind of object oriented fashion are instantly available for all data science needs.
• Mentor juniors . Groomed Unbxd tech content for Berlin Buzzwords - Europe’s premier dev conference (author/company blind review based selection).

Apr 2016 – Mar 2017 · 11 mos · Gurugram, Haryana, India

• Led research and innovation team, products, related strategy.
• Rolled out India's first ( world's third) flight fare prediction engine (buy now or wait recommendations) v1.0 - product design, mathematical modelling, parts of dev leading small team. Predictions are live since Jan 2018 and run at a sustained average accuracy of near 90% saving millions on Indian air travellers wallets and has successfully predicted Indian airline sale prices too .

Jan 2015 – Feb 2016 · 1 yr 1 mo · Greater Nice Metropolitan Area

• Following the Ignite Program at Stanford worked on a new venture that would automatically generate building floor plans from building design requirements. The venture had made its way to being selected as one of the top entries in the 2014 Intra-Ignite group venture competition.

Sep 2011 – Jul 2014 · 2 yrs 10 mos · Paris

• Responsible for Appsfire's server infrastructure, related decisions(scalability/hire(vendor decisions/technical specs)cost-reduction/capacity planning). Designed and implemented critical infra monitoring/alerting and mysql tools. Hands on work - devops using Linux/bash and in close collaboration with Appsfire team for mission critical operations, and infra performance improvement ( diagnosing production issues and fixing them) and scale up. Also: dev/support of Appsfire Android app/engagement SDK: major revamp of legacy android code shipping 2 releases, architected v1 of Appsfire's android ad-unit engine - both device side and web services side(php)

Sep 2005 – Sep 2011 · 6 yrs

• 1. Worked on a method of semantic integration( subtopic co-location) of multiple web documents on the same topic
• 2. Built a big data-search convergence platform that ETL-ed millions of raw airline ticketing records, created serialized database internal tree like structures of airline domain search keys and pre-computed statistical distributions as values serving the distributions from memory. It had applications in several divisions at Amadeus in consumer and corporate travel
• 3. Developed (full-stack) a suite of applications leveraging the above platform (Java/ J2EE/javascript/Highcharts) and data. 4 apps on worldwide airline ticketing data and 1 app for Amadeus E-Commerce Quality Control Infrastructure
• 4. Built a data dependent load distribution algorithm for parallel low airfare search (Master Pricer Calendar) that improved response time by 30% on average while improving airfare recommendations.
• 5. Managed collaborative projects between Amadeus and external partners: Airline Industry Consortium(MIT), French govt initiative(Pôles de Compétitivité) to bring startups, research institutions(INRIA) and large companies together.

Apr 2001 – Sep 2005 · 4 yrs 5 mos

• Design and Implementation of automatic update software, high throughput airline ticketing middleware server programs and automatic stress testing tools.

Mar 2000 – Apr 2001 · 1 yr 1 mo

• Responsible for engineering and managing aspects of the HCL Presenter application development suite ( used by some major fortune50's of the world) in collaboration with Blaze Software, California. Before java, this was
• the way to achieve platform independence with C/++. You could
• write your app for example on a Windows machine, take the same code base to a different platform e.g. a green screen mainframe, recompile the binaries
• and deploy them there, without any changes. 39 combi (hardware-OS-compiler) platforms were supported.

Sep 1996 – Dec 1999 · 3 yrs 3 mos

• Designed and implemented parts of the University's student administration system.

Mar 1996 – Dec 1999 · 3 yrs 9 mos

• Automated learning of search heuristics for combinatorial search problems in engineering design.

Feb 1994 – Nov 1994 · 9 mos

• Responsible for IT infrastructure and decision support environment in the construction project planning and monitoring domain.