Gulshan Kumar

Led a cross-functional team comprising 2 ML Engineers and 3 Data Engineers to enhance Loco platform recommendations. Establish data-driven objectives with measurable key performance indicators (KPIs) to drive a targeted impact on user engagement and revenue. Designed and developed a real-time NSFW content removal model to enhance user safety and experience on the platform. Implemented a fraud detection model, yielding a ~80% decrease in emulator viewers and a 60% reduction in payment-related fraud, resulting in enhanced financial security and increased user confidence. Revamped a high-performing sticker generation model utilizing advanced machine learning, with a 90% accuracy in creating visually appealing and contextually relevant stickers.

Oversaw end-to-end product development lifecycle, from ideation to deployment, ensuring timely and high-quality delivery of AI-driven clip generation from videos while maintaining the context. Directed the development of proprietary AI algorithms for video processing, contributing to a 25% increase in Clips content engagement. Designed and implemented a scalable and robust technical architecture, optimizing video processing workflows and reducing latency by 27%. Led a team of 7 interns to develop the complete model for brevids with accuracy of approx ~95%.

Implemented Conversational Ai Assistant backend using Rasa framework in python with a pipeline comprising Intent Classifier, Spacy, CRF Entity Extractor, POS tagging, and Constituent Parsing. Led the development and deployment of cosmos chatbot. Led up-gradation of the architecture of the chatbot and worked on developing a strategy for expanding the chatbot to new use-cases. Up-gradation of architecture led to saving infrastructure costs of $13,000 annually. Led research on intent detection and named entity recognition approaches for chatbot (LUIS, BERT, quantized and compressed versions of BERT).

Developed a scalable chatbot named ask-Finance, the pioneering chatbot for AEXP Finance end-users, catering specifically to end-users from Oracle GFO and Operational Excellence teams. Implemented as an entirely homegrown low-latency microservices-based polyglot application, It is deployed on the Amex enterprise cloud platform, achieving an impressive accuracy rate of 96%. Optimized the intent identification and entity recognition model, accomplished an accuracy of 98%. Introduced a selenium script for automation testing which led to saving 20 hrs of time per month on average. Collaborated with multiple teams to provide a GDHA architecture for ABP tracker.

Winner of Goldman Sachs Quantify 2018. Runner-up in American Express Artificial Intelligence challenge 2018