Jul 2024 – Present · 1 yr 8 mos

• Product Innovation & Delivery: Led development and release of several AI-powered solutions in Marketplace including RAG-optimized QnA assistant (Wayfinder), multi-agents root cause analysis (ALFA), hierarchical multi-label classifier (SmartCase) & LLM topic generation service (Reviews Manager) to enhance seller support experience.
• Business Impact: Drove measurable improvements across Marketplace metrics: 10% increase in category selection accuracy, 42% reduction in case creation time, 33% growth in seller NPS, and 30 FTE savings through operations optimization
• Technical Leadership: Spearheaded MLOps excellence through Dockerized CI/CD pipelines, automated changelog generation, and OpenAPI-first development while standardizing engineering practices for scaling AI products
• People Leadership: Achieved 100% score on Associate Engagement survey (13% above benchmark) leading 9 member team(7 DS, 2 MLE team). Hired 5 new scientist Staff-level engineers and established mentorship program for each.
• Thought Leadership: Authored white papers on AI agents, productionized novel multi-agent solution (ALFA) for automated root cause analysis, and led cross-functional AOP planning for emerging technologies

May 2021 – Jul 2024 · 3 yrs 2 mos

• Developed & launched new autonomous self-checkout systems utilizing Computer Vision to recognize products, removing need for customer scanning.
• Improved online metrics including reduced average transaction times & enhanced customer experience.
• Customized & fine-tuned multi-modal OpenAI CLIP via knowledge distillation to enable training with <90 images per class leading resulting in 9.2% improvement in precision at same recall.
• Developed automated picklist generation system using graph based Louvain Community Detection and LLM-based attribute extraction to
• find visually similar looking items at scale reducing picklist generation time from from 4-5 days to <1 day for 7500 Walmart products.
• Drove strategy collaboration for rollout across product, business, engineering.
• Customized & fine-tuned multi-modal OpenAI CLIP model via knowledge distillation to enable training with small datasets.
• Engineered auto-annotation platform with unsupervised machine learning techniques, increasing annotation throughput by 8x.
• Independently developed customer-facing picklist feature to address highly similar looking products, increasing system precision.
• Created production-grade, highly scalable training pipelines leveraging Kubernetes, Dataflow & GCP to train hundreds of models.
• Led development of hybrid visual & textual product recognition algorithms leveraging deep learning to accurately classify over 100,000 products in 5,000 stores on Jetson AGX edge devices.

Aug 2019 – May 2021 · 1 yr 9 mos

• Project: Anomaly Detection on Transactional logs
• Leading Data Science team to build a generic and automated unsupervised Anomaly Detection platform for POS systems.
• Developed a fully scalable & distributed pipeline on Google Cloud (GCP) with 6 key Algorithmic Modules & 3 others in progress to make this fully automated.
• Algorithmic Modules include semi-automated KPI identifier, weighted multi-algorithm Engine, scoring mechanism, UI based validator, alert reduction mechanisms, signal classifier, seasonality module and feedback learning.
• Each algorithmic module above consists of a group of machine learning algorithms or statistical methods designed to solve problems like False positives, seasonality, model selection, high number of alerts, scale, scoring, relative anomalies.
• Project: Distribution Channel Recommendations for pharmacy drugs
• Leading a 6-member team to design and implement recommendation model which finds optimal distribution channel for drugs.
• The solution is based on 14 factors (for instance: service level, cost, rebates, Inflation, holding cost etc) to maximize benefits.

Sep 2017 – May 2021 · 3 yrs 8 mos

• Summary:
• Purchase Order Forecasting (Forecasting and Optimization)
• Produce Quality Inspection using Computer Vision (Deep Learning, keras, Tensorflow)
• Pricing Estimator tool for Walmart Pharmacy
• Supplier Selection tool for Walmart Pharmacy
• Details:
• Purchase Order Forecasting (Forecasting and Optimization)
• 1. Build a forecasting engine to provide 13 weeks out forecast to 9000+ Walmart suppliers resulting in estimated hundred of millions of dollars saving for Walmart and better planning, transportation and scheduling for suppliers.
• 2. Combined time series, regression & convex ensembles to choose best model for each item/dc.
• 3. Led a team and coordinated with the business, product & offshore to ensure successful product delivery.
• Produce Quality Inspection using Computer Vision (Deep Learning, keras, Tensorflow)
• 1. Automated produce quality inspection in the DC’s using deep learning methodology removing human bias & saving labor time with an estimated saving of $8-$10 million annually
• 2. Identify defect types & defect severity in produce items using advanced object detection and object classification
• 3. Worked with a team of Data Scientist to develop advanced object detection and classification algorithms with >92% accuracy to identify defect type in the produce item.
• 4. Built algorithm on TensorFlow backend with NVIDIA GPU’s to reduce training from days to hours & prod latency
• Pricing Estimator tool for Walmart Pharmacy
• 1. Building version 2 of the pricing estimator tool to provide an estimate cost of buying prescribed drugs from Walmart pharmacy beforehand.
• 2. The solution is envisioned to be a combined effort of machine learning, rule based and statistical solution.
• Supplier Selection tool for Walmart Pharmacy
• 1. Build an interactive R shiny application which helps Walmart pharmacy buyers choose best suppliers for based for different drugs.
• 2. The adhoc request execution fulfilment time got reduced from 4 hours to less than 3 minutes.