- I led the design and implementation of the Bedrock Rerank API, which provides a serverless solution that allows customers to use Rerank models in Amazon Bedrock, from both Amazon and 3P (Cohere Rerank 3.5). The solution also integrates with Bedrock KnowledgeBase and allows KB customers to easily integrate with the Rerank API in their RAG workflows to improve accuracy. Get started at https://docs.aws.amazon.com/bedrock/latest/userguide/rerank.html- I led the design and implementation of metadata selection for CSVs in Bedrock KnowledgeBase - a feature that lets you specify structure information about CSVs in corresponding metadata files which is then used by the service to parse and store the CSVs in a special way to allow filtering based on customer-specified columns during retrieval.- I led the design and implementation of CloudFormation (and CDK) support for Bedrock KnowledgeBase.- I was part of the team that launched Knowledge bases for Amazon Bedrock, a Generative AI service that allows customers to bring in their own data sources to enable retrieval augmented generation and use, at inference time, data that was not used during a foundation model's pretraining.- I led the design and launch of Assisted Slot Resolution for Amazon Lex, a new feature that uses Generative AI to improve the resolution of supported built-in slots. The feature was announced in re:Invent 2023 as part of Adam Selipsky's keynote- I was part of the team that developed the Visual Conversation Builder for Amazon Lex, a drag and drop experience for building bots

- I was part of the team that worked on Lex V2, a revamped conversational AI experience that became generally available in January 2021. - I worked on the launch of Lex V2's import/export feature which allows Lex V2 customers to easily import and export bots.- As part of the Comprehend Medical team, I worked on setting up machine learning pipelines to productionize the backend models of the InferSNOMEDCT API.- I worked on launching support for GRXML slot types in Lex V2 and was responsible for designing and implementing the APIs for the same.

- I designed and implemented a failure handling mechanism for one of the microservices of AWS Lex to allow automatic retries of failed requests, without any manual intervention- Helped reduce operations burden of the team and received a return offer as a result of the success of this project

During my time at Visa, I was part of the Network Processing team where:- Created a report comparison tool in Python for an existing application that reduced comparison time from 30 minutes to a few seconds- Worked on designing & creating an automation framework for a set of internal applications using Java- Helped in developing the continuous integration system for a set of projects using Jenkins - Designed & created the POC for an application to replace existing Mainframe-based system, using AngularJS- Successfully built an Android app for blocking cards during one of Visa’s internal hackathons

- Was part of the Bing team where I worked on extracting attribute-value pairs from product descriptions taken from retailer websites like HomeDepot- Designed and implemented a unique algorithm employing Natural Language Processing techniques: I extracted the relevant phrases using Dependency Parsing along with certain domain specific rules- Achieved an accuracy of greater than 70% - this method was later integrated into Bing search- Used C#, Microsoft’s internal map-reduce framework and the Stanford NLP Parser

- Implemented a basic version of the RSA algorithm using C and the GNU Multi-Precision Arithmetic Library- Was able to successfully perform encryption and decryption of pdf, image, text and doc files using 1024-bit keys