Amit Sowani

About Amit Sowani

Analytic Scientist with 11+ years of experience building analytic products in retail and financial domains. Skilled at model development processes from data visualization to variable generation to model training and evaluation. Accomplished diverse projects in supervised and unsupervised domains using structured and unstructured data sources. Experienced at productizing analytic models and managing a team of scientists for successful delivery of analytic projects.

Experience

1. Analytics Science Senior Manager

   FICO

   Jul 2008 — Present · Bangalore, IN

   Retail Action Manager (RAM) – Personalized Offer Recommendation Engine• Developed a model factory which predicts customer purchase propensity, based on their historic data.• Built tools for automated variable generation using genetic algorithms and automated scorecard model training with checks to avoid overfitting.• Parallelized processes using Hadoop Map Reduce framework for scalability.• Deployed the product for large North American retailers with millions of customers and thousands of offers.Streaming Analytics – Real time transaction analytics • Designed architecture based on Storm which processes transaction history, summarizes profiles and outputs scores based on an underlying model. • Implemented recursive profiling to perform low latency update of profiles using existing profiles stored on Cassandra.• Trained logistic regression models which incorporated Latent Dirichlet Allocation (LDA) scores to account for cross product interactions.• Developed a self learning feedback loop with automated training dataset generation, model training and an A/B testing framework• Ported RAM clients to enable recommendation of highly contextual offers.Expense Report Navigator – Travel and Expense abuse detection• Developed unsupervised outlier detection models using Minimum Value Ellipsoid (MVE) model as an enhancement over Principal Component Analysis (PCA) based outlier detection.• Designed self-calibrating models to adapt to evolving abuse patterns using quantile estimation techniques as inputs to the model.Lifecycle Analytics – Originations, Customer and Debt Management • Developed a library of characteristics based on bureau data using a templatized approach characterized by the financial entity, aggregation function and a time window • Built model development pipelines in Python for various customer lifecycle use cases like loan origination, customer retention and payment projection.• Explored explainable AI techniques to explain black-box models.

Education

• Indian Institute of Technology, Bombay

  Bachelor of Technology (BTech)

  2002 — 2007

Skills

• Analytics
• Business Strategy
• Management
• Strategy
• Business Analysis
• Data Analysis