- Team Management: Managing a global team of 6+ ML Engineers and Scientists across the US, India, and Ireland, fostering collaboration and driving innovation in ML projects.- ML Projects: Spearheaded the development of a content restructuring model (and improved product knowledge graph) using deep learning (using computer vision, NLP) and Google Gemini (LLMs + RAG), making it easy for our partners to onboard new SKUs. Reduced duplicate products by over 50% with embedding-based similarity and pHash, boosting customer engagement. Championed usage of LVLMs (BLIP, Gemini Flash, ChatGPT) for product attribute prediction, leading to improve search results and engagement. Used deep learning (neural networks) to identify similar items and identify outliers for making prices competitive. Deployment on GCP and on-prem using Docker and Airflow.- Product Discovery & Recommendations: Championed the use of in-house recommendation (combination of matrix factorization, collaborative filtering, aesthetic similarity, association rule learning) algorithms on new business verticals - Overstock and Zulily, achieving over $2M in cost savings. Directed the 'Shop the Room' project using object detection using Segment Anything Model (SAM), LVLMs (Gemini-Flash, ChatGPT) for personalized recommendations, resulting in higher average order sizes.- Stakeholder Engagement: Engaged with multiple stakeholders to identify new ML use cases, aligning the ML product vision with business goals to drive impact.

- Team Lead: Managed a team to build ML solutions to fight fraud, automate claims - leverage AWS and deep learning (computer vision, natural language processing) for insurance industry. Product built: https://www.verisk.com/products/digital-media-forensics/- ML Solutions: Applied state-of-the-art computer vision techniques—image classification, object detection, and instance segmentation built in Tensforflow/ Pytorch —to build products to detect fraudulent claims. Set the strategic vision for the image & document editing detection product, aligning development efforts with stakeholder expectations.

- Built ATHENA (PySpark), an automated tool for building and tuning machine/ deep learning models text data, saving hundreds of hours on model building- Built a algorithm (machine learning + sentence similarity) to tag unlabeled text bypassing human annotation to generate larger corpus for training machine learning and deep learning models- Built and maintaining self-learning machine learning (logisitc regression) and deep learning (charCNN) pipelines in production- Built pipeline to work with non-English data within the current setup using Google Translate- Use PySpark to distribute extensive computations (ml models + grid search + generating new data)

- Lead setup of cross-device attribution and user journey; exploring use of Markov models to credit sales to device - mobile & Desktop- Developing logic for predictive maintenance of air filters for timely replacement using pattern mining- Developed Tableau dashboard & algorithm for early detection of non-performing ad sets on Facebook

- Built a scalable system (using multiprocessing in Python) to find similarity between thousands of documents using difflib Sequence Matcher/ Levenstein Distance /cosine similarity/ word embeddings generated by word2vec & gloveCode at https://github.com/analyticsbot/document-similarity- Analyzed Stackoverflow data to understand the factors that determine a user’s reputation Code at https://github.com/analyticsbot/stackoverflow

- Worked with Marketing to analyze/predict participant loan behavior in the three-to-six month period- Worked closely with the marketing team to add digital capability ensuring data driven decisions- Leveraged click stream data from www.netbenefits.com to improve participant experience and marketing effectiveness. Used D3.js for the senior management to demystify participant web journeys, adding efficiency in decision making- Analyzed and optimized Fidelity's 401k related marketing campaigns. I was part of a team tasked with understanding our participant's journeys and digital activity, helping participants to save more in their 401(k) plans- Conducted Python training sessions for colleagues, adding new capability

-Implemented an user-preference based recommendation engine product - JARVIS (Java, Mahout, Hadoop, Spark, Cassandra, RabbitMQ), now one of four flagship products of the company - Team Received nomination for "Spirit of LatentView Award".- Created internal social media analysis product - SocioBOT (Python), after understanding the problems faced by in-house teams, saving hundreds of hours of manual work.- Developed a clustering approach for path-to-purchase analysis to identify high frequency path sequence and high drop off pages for an e-commerce client (Python, D3.js, SQL, Excel).- Developed several executive-level Tableau dashboards that integrated forecasts and business performance metrics contained in various SQL tables for effective data visualization (Tableau).- Web scraped multiple review websites, performed text mining to extract sentiment to identify innovation and purchase driver indicators and supervised creation of Tableau dashboards for senior management for a US retail client (Python, Scrapy).- Planned, organized, and executed a company-wide Hackathon, involving more than 200 participants, to push the interaction of new technologies and paving way for future hackathons.

- Implemented a predictive model to improve efficiency of ZipDial’s online marketing campaigns by predicting the customer response.- Offered strategic & tactical insights to clients by developing excel dashboards on using campaign data.- Involved in End-to-end Project management including requirement analysis and client management.