Srividhya Narayanan Email & Phone Number

As part of this project, built multiple proof of concepts like Customer Service Rep assistant, Marketing editor’s assistant. Presently, client is aspiring to build an assistant for policy research SME’s leveraging generative AI. The app needs to scan all payer sites (like UHC, Aetna etc.) and extract relevant information, compare with client policies and generate findings.● Built the proof of concept in AWS, which was demoed to the clients and paved way for implementation● Currently working on implementing the application in client’s environment for production rollout● Hands-on experience in leveraging Bedrock Claude models, FAISS vector storage, Lang chain, selenium web scrapers to build the app● Currently playing the role of AWS architect to implement the app as a pilot for the client’s policy research team

Client wanted to explore the possibility of using LLM’s to detect missing appliance parts on the shelf, enabling seamless parts ordering. The same functionality can also be used to detect new parts that need to be ordered by analyzing the images of the appliances.● Leveraged Gemini Pro and Flash and dynamically created few-shot image prompts to detect the defect in the image or the new part photographed● Performed image preprocessing including format conversions, glare removal etc. to improve accuracy● Explored the use of Auto ML in Vertex AI for the same problem and compared the pros/cons against Gemini● Explored use of multi-modal RAG to search and identify images that can be passed in the prompt for few shot prompting● Designed the GCP production architecture for the solution to be implemented in production

The client faced the issue of lack of metadata insight to answer some of the questions from data analysts/developers. Client wanted to build a metadata assistant that can answer questions related to metadata from the Big Query backend tables. It was also extended to analyze big query logs and answer data extraction related real-time questions● Designed the GCP architecture to use Gemini Code-bison and convert the user’s natural question into a Big Query SQL● Leveraged BQ user defined functions to implement Levenshtein distance algorithm that can be used by the SQL to sort the most relevant results.● Built the chatbot with the development team to answer user questions and also context across conversations

SIP is an internal product at Cognizant for standardizing interactions between payers and providers for validation and trial purposes. SIP had a requirement to automate provider data ingestion to fast-track onboarding. The idea was to read the 1572 documents and extract the provider information that can be ingested into SIP● Leveraged AWS Textract to read 1572 Federal PDF documents and extract information● Integrated with the application to invoke the endpoints to parse 1572 documents and prefill provider information● Worked with AWS technical team to resolve issues faced around entity relationship features of Textract

As AI architect, built the components to create Alexa Skills and Google Dialog flow Actions to enable remote execution of commands to the car (e.g.: Lock my car). Led the pioneer efforts to implement name less invocation (CVSA) for cars, by partnering with AWS and Nissan teams. Technology: AWS Lambda, Alexa skill kit, GoogleDialog flow, Google Functions, Google Conversation Actions, Node JS, Type Script● Implemented name less invocation in production in Oct 2022● Implemented additional commands like seat heater control, defrost controls to Alexa● Built the design for app to app linking with Nissan Apps● Optimized design to adhere to the strict performance SLA’s and enable timely response to Alexa user