Prof. Sarah Preum, Persist Lab - Leveraging Large Language Models for custom text extraction and filtering to improve textual data annotation - Employing the extracted text to improve downstream text sentiment/stance detection task’s accuracy by 20% - Evaluated scope of LLMs(GPT-4, GPT-3.5, Llama2-7B) for mining emerging opinions in online health discourseProf. Yaoqing Yang in colab w Mahoney Lab(UC Berkeley) - Working on visualizing high-dimensional loss functions - Studying the different phases in model learning and understanding relationships between local and global metrics

• Improved the query response from LLMs for tabular data using modern prompting techniques by 10%• Fine-tuned OpenAI/Llama2 models to respond from the custom client data only to further improve responses

- Worked along with the Professor for grading and creating solutions to problem sets.- Cleared doubts of students to improve their direction of approach in the subject.

- Architected and implemented large-scale cross-pod features catering to 20M daily active users- Implemented lambda function to serve requests from AWS tg directly reducing the latency of the api by 50%- Owned the Meesho product-over-web backend service and improved its structure to make it scalable- Developed an internal automation tool to find the respective team POC, saving time for the entire organization

- Created face-recognition module recognizing faces from the documents and live video using OpenCV- Developed the preprocessing module for the document-engine which improved text extraction by 20%