---- AWS ---- Solving novel optimization problems using machine learning.---- Alexa ----Building cool MultiModal experiences in Amazon Alexa. Dealing with 100,000+ TPS.Creating LLM-based tools using RAG and Agentic flows.

1. Built an end-to-end machine learning platform for acoustic image data, including data collection, processing, training, and real-time inference.2. Trained models like ResNet, MobileNet, and Transformers for human body pose, facial expression, and activity recognition using acoustic signals. Utilized data augmentation and masking techniques to createstate-of-the-art systems, achieving 90%+ accuracy on most tasks.3. Secured $240,000 in funding from Cornell University and $50,000 from National Science Foundation.

1. Proposed and developed a machine learning recommendation tool for auto-filling customers’ preferred delivery dates, reducing clicks from 5 to 1 for single-click checkout2. Migrated notifications platform to AWS, implemented an S3-based request and response tracking framework, ensuring timely and accurate notifications.

Part of the VAS(Value Added Services) team. 1. Created a widget for Amazon's mobile homepage that is selectively shown to the customer based on machine learning models. The information on this widget reduced customer anxiety for service orders by 2%, saving millions of dollars spent on customer support.2. Designed and created a caching system to reduce the latency from 300 ms to ~10 ms. This cache's hardware can be upgraded in the future to reduce to latency to sub-millisecond levels.3. Currently Amazon only supports warranties for orders that are fulfilled(shipped) by Amazon. To extend this to orders fulfilled by merchants, I designed a solution which required deep diving in multiple code bases and working with 6 different teams at Amazon.

Amazon is creating and testing a new B2B2C business model that would provide free smart home installations to the customer. Currently, an SDE was generating invoices and this was taking a huge toll on the development process. To shift this load to the business team I created a Fund Settlement Portal that would allow to generate and track the status of the invoices. This thing was built and designed end to end by me.Skills used: Java, CSS, HTML, JavaScript, Amazon DynamoDb, Angular, Ajax, Intellij, lot's of Amazon internal software stack.

Teaching Assistant for Statistics Course.

1. Integral part of the team that developed India's 1st VR platform for the construction Industry, Trezi. 2. Built the cloud stack and integrated it with the VR app for Trezi, which allowed users to download textures and models in real-time. 3. Enabled Trezi to create a multi-million-dollar marketplace by allowing different companies to sell products through VR. Tech used: Amazon Dynamo db, Amazon S3, electron, Unreal engine,coherent UI, AWS c++ sdk, node js, c++ .