Projects
Welcome to my portfolio! I am a Master's student at Purdue, who is passionate about the applications of AI in healthcare and other sectors. Scroll down to learn about some of my projects.
Image Classification using VGG16 with AWS Deployment
Created an end-to-end pipeline for classifying input images based on their landscape. Initially, I started by comparing the accuracy of the pre-trained VGG16 model to a standard CNN. After deciding to use VGG16 for the project, I created my application backend using Flask, with Docker and Kubernetes for containerization and autoscaling respectively. I deployed the application to AWS SageMaker using ECR to store my Docker image and Lambda functions to transform application inputs as needed.
Hematology Report Analysis and Automation
Private Repository
Developed a Python codebase to parse and interpret the output of a Hematology Analyzer, store the interpreted results in the concerned hospital's PostgreSQL database and automate the generation of human-readable reports. Tested and deployed the project to a remote PC using RustDesk.
Hybrid RAG-based AI Chatbot
After scraping data from a local university's webpage, I used a the "Llama 3-8B Dolfin" large language model to create an interactive AI chatbot to answer questions about the university's faculty. The project pipeline involved data preprocessing, applying recursive chunking, generating text embeddings and performing prompt engineering. The code was adjusted based on several key considerations, such as the LLM’s context window, the size of the generated embedding vectors, and the format of the input data. Later, I integrated the chatbot with Hybrid RAG architecture to ensure that embeddings were matched based on a hybrid score generated using both cosine similarities and keyword matching.
Purdue ELI Corporate Consulting - Vanilla RAG vs. MRAG
(In Progress)
Private Repository
Working to integrate client's existing Vanilla RAG-based AI chatbot with Multi-Headed RAG architecture. This will enable intelligent decomposition of the query into sub-queries that can be answered individually before being combined to form a single, comprehensive answer. The process involves integrating the MRAG approach, performing AB Testing to compare it with the existing RAG infrastructure, determining how best it can be deployed, and providing the client with consultation about whether to move forward with this new idea.
