Jul 2025 – Present · 8 mos

• Multimodal GenAI
• ASR (Automatic Speech Recognition)
• Agentic AI

May 2022 – Jun 2025 · 3 yrs 1 mo

• Leading the Conversational AI team, delivering advanced GenAI, NLP, and ASR solutions across US and international markets.
• Developed and scaled Walmart’s GenAI and ASR platforms—introducing agentic flows (RAG, API agents), low-cost fine-tuned LLMs, and a proprietary speech-to-text solution that outperformed major cloud providers in both price and accuracy.
• Launched few-shot intent & entity detection models and contextual NLP for chatbots, delivering substantial improvements in accuracy, latency, and cost. Developed MLOPs and analytics for the platform.
• Continued investment in research and innovation charters.

Feb 2021 – Apr 2022 · 1 yr 2 mos

• Lead the Conversational AI Platform - data science team.
• Launched hierarchical NLP (label-aware BERT architectures) and proactive personalization models for chat and IVR, achieving iterative gains in intent detection and customer experience.
• Drove successful M&A evaluations and led cross-functional collaborations between AI, product, and business teams.
• Drove innovation within the team, resulting in the publication of research papers and the filing of patents.

Jan 2020 – Jan 2021 · 1 yr

• Leading the 'Conversational AI' data science team.
• Envisioned and built the 'Converse for Customer Care' initiative at Walmart, to use AI for powering seamless conversational experiences for customers.
• Built a team of data scientists and engineers from scratch. Drove the AI strategy, roadmap, strategic partnerships and adoption across Walmart. The applications include NLP and ML solutions deployed for personalized and seamless conversations across channels (Chat, IVR) and domains/markets.

Nov 2017 – Jan 2020 · 2 yrs 2 mos

• Responsible for: a) Leading display retargeting team to scale the program efficiently on the Open Web. b) Leading Voice commerce data science for ‘Walmart.com Customer Care’. c) Hiring and building data science teams for Display and Voice. d) Managing stakeholders across teams and verticals to deliver the projects effectively.
• NLP models for Voice commerce: Developing MaLSTM models for FAQ matching for Walmart voice assistant on smart devices. Creating NLP solutions for Walmart.com customer-service chatbot using SOTA deep learning models like BERT. (Python, Tensorflow)
• Scaling the Display Retargeting initiative: Scaled campaigns by launching and building bid-models suitable for Open web inventory. Designed and launched campaigns for measuring and optimizing incrementality of display ad campaigns. (SparkML, Scala, Python)
• Enhancing user response prediction and recommendation models using deep learning: Generated item embeddings using browsing sequence (Word2Vec) and images (CNNs) for the millions of items in the Walmart.com catalogue for aiding downstream prediction tasks. (SparkML, Scala)

Nov 2015 – Oct 2017 · 1 yr 11 mos

• Responsible for devising user targeting strategy and driving the team towards effective data-centric solutions. Scaled up the channel 50X times at improved ROAS.
• User Response Modeling for Display Retargeting on Facebook(FB): Prototyped and productionized models for click through rate, conversion rate and revenue prediction for retargeting Walmart.com Desktop, Mobile Web and App visitors using Spark XGBoost, RF and LR. Performed extensive feature engineering to extract relevant features from raw site activity. Performed multiple A/B tests around user modeling and bid scaling. Tests against FB-optimized campaigns revealed that our algorithms/models were able to drive 1.5X efficiency at the same scale. The published models drive all Walmart.com retargeting on FB (Millions of $ monthly).(Spark, SparkML, Scala)
• Recommendation Systems for Display Retargeting on Facebook(FB): Built recommendation models for retargeting Walmart.com users using collaborative filtering and decision trees at different stages. The feature set consisted of both user and item attributes. Built an offline Recommender evaluator to evaluate recommendation quality and shortlist only the superior ones for A/B tests using MAP/Recall metrics. Experimented with multiple recommenders and item-templates on FB and adopted the best strategies. (SparkML, Scala)
• Forecasting and Budget allocation: Built models for Revenue Forecasting for the Display channel yielding optimal AdSpend allocation across days and channels using adspend-revenue elasticity curves. Automated Budget-Allocation across multiple Facebook and Criteo Campaigns based on the Revenue Over AdSpend (ROAS) goal by modeling as an optimization problem. (Python, Scipy, Numpy, NLOpt, Pandas)

Jul 2014 – Oct 2015 · 1 yr 3 mos

• Responsible for devising FB Prospecting strategy, optimizing the Walmart.com Comparison Shopping Engines (CSE) setup and building an intelligent log compression tool. Delivered efficient results on all goals.
• Built Prospecting solution for Display advertising on Facebook: Built models for Audience Recommendation, Segmentation and Customer Lifetime Value prediction for acquiring new customers on FB. Devised an algorithm for guaging store-similarity to launch A/B tests on Walmart stores. Developed robust Map-Reduce pipelines for processing user data and used FB APIs to automate the feedback loop for gauging campaign performance. Launched A/B tests with the proposed solutions outperforming FB-lookalike segmentation, yielding ~ 2X ROAS. (Java, Map Reduce, Hadoop, Python, R, Hive)
• Optimized the Comparison Shopping Engines setup for Walmart.com: Item-level Bidding for Comparison Shopping Engines (CSEs) by solving an optimization problem with the objective of maximizing Revenue while also satisfying budget and efficiency constraints. Algorithmically automated matching of Walmart.com Category paths to CSE Partner-Websites' Category Paths using web-scraping. Optimizations resulted in 34% (millions of $) increase in monthly revenue for the channel (Python, R)
• Built an intelligent and distributed Log-Compression tool using pattern recognition. The file sizes reduced to about 5% of the original. (Java)