Mahitha Sangem

Ai Software Developer Intern @Vosyn

ON, CA

Explored fine-tuning and Retrieval Augmented Generation (RAG) implementation with Large Language Models, with given constraints. • Developed a RAG-based Recommender system with structured data for recommending content, based on user history as well as similar user preferences involving collaborative and hybrid filtering.• Performed extensive research on unsupervised clustering techniques and Generative AI-based recommendation systems, LangChain, RAG and LLMs with comparison and benchmarking.• Loaded video data in JSON format, implemented Hugging Face & OpenAI embeddings, ChromaDB vector store & retriever for pulling relevant videos from the database, and OpenAI GPT 3 LLM for delivering the answers to the user query.• Utilized RAG fusion retrieval mechanism for ranking the obtained document chunks based on relevance factors.• Developed a sample chatbot application for retrieving information concerning the user query. Hosted the locally running application on AWS EC2 instance by containerizing it using Docker.• Development using the AWS tools Bedrock, Lambda, API Gateway, SageMaker, Kendra, S3 and DynamoDB. Providing model endpoint API to the front-end team for integration with developed UI.• Utilized Amazon Titan embedding model, FAISS vector store and retriever, Llama and Mistral models as LLMs for summarizing and formatting the responses (from obtained chunks after retrieval) based on the prompt provided to the LLM.• Implemented speaker diarization in multi-speaker scenarios, on GCP Vertex AI with Hugging face. • Keeping track of AI advancements and leveraging best foundation translation models and platforms for developing LLMs.

Carleton University

Master of Engineering - MEng, Computer Technology/Computer Systems Technology

CVR College of Engineering, Hyderabad

Bachelor of Technology - BTech, Electronics and Communications Engineering