Training & Inference of Transformer-Based Ranking Features: Built transformer-based ranking systems for Amazon Search, replacing 50+ legacy ranking features with customer behavior prediction models and enabling large-scale inference pipelines serving 100K+ QPS, contributing to $13B+ revenue impact.

Large-Scale LLM Inference Infrastructure for Search: Architected large-scale LLM inference infrastructure for semantic search and ranking, deploying 1B+ parameter models and enabling high-performance vector retrieval across 260M+ products with low-latency production serving.

Large-Scale LLM Training Infrastructure & MLOps: Developed distributed LLM training platform supporting models up to 32B parameters, enabling scalable experimentation, optimized training throughput, and production deployment for large-scale AI systems.

Discovering Unexplored Amazon Products on Search: Built reinforcement learning–driven discovery systems surfacing new products to 150M+ daily users, driving a 4× increase in new product purchases and improving marketplace exploration.

Conversational AI Shopping Assistant (Amazon Rufus): Built distributed cloud-native ML services powering conversational shopping experiences (Autocomplete, Deals, Adaptive CX) for 310M+ daily active users, processing 250K peak TPS. Developed high-throughput streaming pipelines for conversational AI events and improved LLM-powered query coverage to 96% with sub-20ms latency.

Recommendations & Related Search (Search Q/A Widget): Launched generative-AI powered recommendation experiences within search Q&A, automating product discovery for shopping queries and scaling to 100M+ daily active users. Led experimentation and A/B testing to optimize engagement, recommendation quality, and business metrics.

• Centralized Data Extraction Library: Developed an enterprise-wide data platform to perform financial analysis and publish reports to the clients; providing fast, reliable, and convenient way to access data from various sources.

• Filings Data Extraction Service: Built NLP-powered data retrieval system extracting key sections from multilingual financial filings reducing research time by 10x via RESTful API and react-based search interface.

• Financial Time Series Anomaly Detection: Improved financial data quality from 96.4% to 99.9% through

ML-powered anomaly detection pipeline leveraging statistical, supervised learning, and clustering techniques.

• MSCI’s NLP based Thematic Index: Spearheaded ML-powered financial index generation, enabling expansion to 15+ new investment themes. Built end-to-end NLP pipeline to generate deep learning based probabilistic equity classification, reducing research time by 6 months per theme.

Independently involved in the end-end development of the following modules –

1. Chat module - Text communicator for two or a more users

2. Notifications - Event based application alerts to the user

3. Data Visualizations - Report data driven visualizations

4. Selenium based web crawler for data extraction from the web

The application is being used my more than 15 hospitals including Fortis Healthcare.

Tech Stack: MongoDB, NodeJS, ExpressJS, AngularJS, React Native, HTML, CSS, JavaScript, jQuery, Selenium, Bootstrap, AWS S3, PhantomJS