Feb 2024 – Jan 2026 · 1 yr 11 mos

• Driving the implementation of advanced ML Engineering and AI solutions to expand capabilities across credit risk practices.
• ➤ Financial Loss Prediction System: Led the development of fully functional production-grade scalable libraries and APIs with numerous configurable parameters for predicting losses for large loan portfolios.
• Developed deep learning time-series models employing Lag-llama (state-of-the-art research) and temporal fusion transformers. Built scalable engineering pipelines and inference libraries reducing forecasting latency by 300% and improving prediction accuracy.
• ➤ Conceptualizing and building ML, AI tools and infrastructure to enhance credit risk underwriting and fraud data science methods used by data scientists at 2nd Order Solutions.
• ➤ Communicating complex technical topics, analyses and data insights with non-technical audiences and senior leadership.
• ➤ Built mentor-mentee relationships to upskill and elevate team performance. Educating junior data scientists on advanced ML and engineering practices.

Aug 2021 – Feb 2024 · 2 yrs 6 mos

• ➤ BNPL Risk Prediction Model: Led the development of a classification model predicting default probability for a FinTech's Buy Now Pay Later product using gradient boosting machines (XGBoost).
• Reduced portfolio risk, 82% classification AUC, outperforming existing models and reducing acquisition risk by 22%.
• Developed production ready solution including data engineering and model inference pipelines, drift capturing modules and model monitoring dashboards.
• ➤ Auto Risk Scoring Model: Built a machine learning model (Gradient Boosted Tees) to predict default probability for an auto-lender resulting in 16% reduction in portfolio risk and generating $35 million in cost savings.
• ➤ Data Pipelines and Feature Evaluation System: Developed data engineering pipelines and feature evaluation systems at scale for ML risk models for a top 3 US bank.
• Reduced feature testing-to-selection lag from several days to a few hours impacting 3 large modeling teams. Implemented Spark based hypothesis tests in PySpark and SparkML libraries.
• ➤ Document Verification Service API: Developed a streamlit application employing Generative AI LLMs to automate document verification for small business loans. Leveraged Retrieval Augmented Generation (RAG) architecture on GPT-4 using Llamaindex for engineering pipelines.

Aug 2020 – Jul 2021 · 11 mos

• ➤ Automated Modelling Package API: Built a python ML software package for automating model builds, feature selection, transformation, and diagnostics. Reduced model exploration timelines from several months to few weeks. Package is now widely used by Data Scientists in the company.
• ➤ Built data science algorithms for auto-verifying credit applicants' income using statistical techniques and Bayesian methods such as KDEs, Monte-Carlo simulations, and hypothesis tests. Resulted in operational cost savings of $18 MM per year.
• ➤ Built model drift capturing system for customer acquisition and engagement models. Stress tested features and provided statistically based recommendations for re-entry and credit line increase strategies for a top 3 US bank.

Nov 2019 – May 2020 · 6 mos · Charlottesville, Virginia Area

• ➤ Object Detection for Highway Tolling: Published a paper on building an automated vehicle detection system for highway toll cameras using object detection and classification with ML algorithms for computer vision. Utilized YOLOv3 on PyTorch. [https://ieeexplore.ieee.org/abstract/document/9106682]
• ➤ Geo-Recommendation Engine: Developed a location-based recommendation system using geo-graphs data and prediction engine using graph representation learning and recent papers on Neural Graph Collaborative Filtering and Bipartite Network Embeddings.
• ➤ Poster Image Analysis: Conducted multilabel classification of genres from movie poster images using CNN based ML models like ResNet. Explored the impact of poster design on user perception and box office performance. Implemented models and libraries with FastAI.

Oct 2018 – May 2019 · 7 mos

• Hopscotch was a $100 MM high growth e-commerce startup when I worked there. Responsibilities included inventory and product segmentation algorithms, tracking, and simulating inventory
• consumption, and delivering statistical insights to procurement and merchandising teams.
• ➤ Inventory Forecasting System: Led the development of the analytical brain behind the inventory reordering system. Built subsystems for inventory forecasting and
• product stack ranking algorithm.
• Involved developing time series models (ARIMA) that automated the management of USD $50M worth of inventory across 50k products.
• Utilized over 100 merchandise features to build a product stack ranking algorithm for use by the merchandising teams to optimize assortment size and identify
• freshness gaps across a catalog of over 50K products. Used principal component analysis and product lifecycles to rank and segment performance categories.
• Resulted in reduced inventory wastage while seeing 100% YoY revenue growth and 120% inventory turn increase in a high growth startup environment.
• ➤ Product Engagement Segmentation: used hierarchical clustering on dynamic time warped dissimilarities of click stream data to classify products by user engagement
• potential revealing the need for a new product listing page sort algorithm.

Jun 2017 – Sep 2018 · 1 yr 3 mos

• Built Inventory Prediction Systems using time series models.
• Optimized product assortment using unsupervised machine learning methods.

Feb 2016 – Jun 2017 · 1 yr 4 mos · Gurgaon, India

• ➤ Modeled performance drivers using ML models (Random Forests, GBMs, XGBoost) to statistically strategize retail category improvements.
• ➤ Built optimization algorithms to maximize store sales by restructuring store layouts. Used Travelling Salesman Problem approach to optimize customer traversal paths, resulting in sales uplifts of 1-7% across different regions.
• ➤ Conducted market basket analysis and statistical experimentation to optimize assortment flow driving positive sales uplift across 28+ categories.

Aug 2014 – Jan 2016 · 1 yr 5 mos · Noida Area, India

• KPI optimization using statistical methods for a large health care firm.

Jun 2013 – Jul 2013 · 1 mo · Mumbai Area, India

• Data Replication on ORACLE server.