Jul 2018 – Jun 2023 · 4 yrs 11 mos · Kolkata Area, India · Hybrid

• Business Data mining: A course on applied AI/ML techniques for the Post Graduate Diploma in Business Analytics
• Analytics in Practice: A course on analytics for MBA students in non‑analytics disciplines

Oct 2017 – May 2019 · 1 yr 7 mos · Bengaluru Area, India

• Managed an AI lab tasked with delivering innovative AI/ML solutions for a variety of business verticals. Selected projects include:
• Store location: Devised an innovative algorithm to identify the best location for a new store, for an Indian furniture chain. The work done here was presented at the International Conference of Business Analytics and intelligence (BAICONF) and won the Best Paper Award.
• Empty container repositioning system: An OR-based solution for repositioning empty containers to exploit upcoming demand at various ports, for an Indian shipping company. The work done here was presented at the International Conference of Business Analytics and intelligence (BAICONF).
• Failure demand analysis: An NLP solution based on call centre transcripts for a Philippines-based insurance company, and identified major causes and trends in customer complaints, s that appropriate policy decisions could be taken.

Aug 2014 – Aug 2017 · 3 yrs · Bengaluru Area, India

• Led data science initiatives to design and deliver solutions for a variety of airline problems. Examples include:
• Segmentation: Used clustering techniques (including some novel ones designed in-house) to find distinct groups of similar entities (PNRs, customers, markets/locations), and defined use cases that demonstrate business value of the same.
• Customer-centric retailing: Helped airlines determine how to offer the right product to the right customer at the right price, taking into account individual preferences, capacity constrained perishable inventory and competitive pressures. Solution components included segmentation, product bundling, price testing, revenue optimization and shopping session management.
• Fare rule design: Designed fare rules that best represented customer preferences, based on traveler data across airlines. The solution involved the use of a novel tree-based clustering algorithm.
• Responsibilities included interacting with internal solution teams as well as airline clients , improving Sabre's thought leadership position in the industry through papers/presentations, and mentoring colleagues on applied machine learning.
• Technical challenges:
• Dealing with data size and diversity
• Building and validating models in a changing environment
• Dealing with imbalanced datasets
• Identifying/articulating the value proposition of ML techniques in a variety of airline solutions
• Synthesizing solution pipelines that involve supervised, unsupervised and reinforcement learning techniques, recommender systems, topic models, multi-criteria decision making (e.g. TOPSIS), genetic algorithms, mathematical programming etc.
• Co-developed variants of clustering algorithms to deal with scale and diversity of data types, e.g. mini-batch variant of $k$-modes for large categorical datasets, evidence accumulation-based hierarchical clustering for large datasets, a novel tree-based clustering algorithm for hierarchical, interpretable clusters.

Jan 2003 – Jun 2014 · 11 yrs 5 mos · Bangalore, India

• Career growth: Risk Technologist (2003-04), Research Scientist (2004-07), Lead Scientist (2007-11), Senior Scientist (2011-2014)
• OUTLINE OF SELECTED PROJECTS:
• Marketing optimization for consumer finance, GE Capital: Project lead
• Developed a suite of applications for CRM decision making in a consumer finance business.
• Key applications included an optimization-based approach for targeting in cross-sell campaigns, and a segmentation-based platform for profiling and targeting credit cardholders for CRM actions.
• Financial impact in excess of $100 MM in asset growth and $20 MM in incremental profit.
• Significant technical contributions include: A genetic algorithm with a novel elitist scheme for predictive modeling in changing environments. A fuzzy mathematical programming approach for robust cross-sell decisions based on uncertain estimates of customer behavior.
• Prognostics and Health Management, GE Power & Water: Project lead
• Developed a performance monitoring system and a diagnostic framework for steam turbines, that combines diverse fault analysis modules using the Dempster-Shafer Theory of Evidence.
• Developed a prognostic model for gas turbine faults using an optimized ensemble of multi-instance learning algorithms.
• Medical image analysis, GE Healthcare: Machine learning expert
• Designed an SVM-based ensemble model as part of a screening application for Pneumoconiosis using chest X-Ray images, for Govt. of China.
• Designed a machine learning model for prescribing oblique scan planes in cardiac MRI exams.

Apr 1998 – Jun 1998 · 2 mos

• Prepared a Request for Proposal and devised a roadmap for implementation of a company-wide intranet, to be spread over 45 locations in 7 states.