Projects

Anomaly Detection in Business Processes using Generative AI

ISB Capstone Project with Deloitte

I led the development of an application aimed at generating actionable insights in expansive business datasets to elevate financial governance. This project encompassed all phases, from liaising with sponsors and faculty mentors, high-level design (HLD) and low-level design (LLD), complete code implementation, large language model (LLM) and agent identification, and testing, to data orchestration and deployment. The result was an AI-powered solution that enhances anomaly detection in business processes using generative AI techniques.

Agentic RAG System for Multimodal PDF Querying.

Independent Project

Developed an Agentic RAG System that integrates LangChain, OpenAI, Cohere Rerank, and ChromaDB to retrieve and synthesize multimodal content (text, images, tables) from multiple PDFs into context-aware responses, and conducted comprehensive evaluations on its accuracy, relevance, and fluency.

Virtual Try-On Prototype with Flask, Twilio, and Gradio’s API using IDM-VTON Model.

Independent Project

Built a virtual try-on application where users can send images via WhatsApp to virtually try on garments, leveraging Nymbo’s Virtual Try-On Gradio’s API which uses IDM-VTON. Integrated Flask to manage the backend API, Twilio for WhatsApp-based interaction, and Ngrok for local server exposure. The system processes the user’s images and returns a virtual try-on result in real-time through WhatsApp.

AI-Powered Personal Browser History Analysis

Independent Project

In this project, I designed and implemented an AI-powered application for personal browser history analysis. Utilizing Python, LIDA, and OpenAI GPT models, I categorized browsing data and offered insights through AI-driven visualizations. The application features an interactive chatbot for custom queries, providing users with a comprehensive analysis of their browsing behavior.

Forecasting Electricity Consumption using Azure Auto ML

Independent Project

I developed a forecasting model using Azure Auto ML to predict electricity consumption in industrial and commercial sectors. The model achieved less than 2% deviation and was deployed as an application on Streamlit. This project demonstrated the practical application of machine learning in energy management and forecasting.

Image Segmentation on Pathology Images using UNet with ResNet

Independent Project

In this project, I finetuned a UNet model with ResNet as the backbone for image segmentation on pathology images. The model was trained with mask images where nuclei were marked, achieving a best validation loss of 10%. This work contributed to advancements in medical image analysis and segmentation.

Building a RAG System for Data Retrieval and Content Generation

Independent Project

I built a Retrieval-Augmented Generation (RAG) system for evaluating in-house data retrieval and content generation capabilities using established Generative AI models like BERTScore and GPT-4. This project involved creating a Knowledge Repository from sentiment datasets, enhancing the system’s ability to provide accurate and contextually relevant information.

Comprehensive Portfolio Management Report

Independent Project

I developed a comprehensive report on portfolio management, which included client profiling, stock selection using AI, and stock performance analysis. This report adhered to a structured format and spanned 26 double-spaced pages, demonstrating the practical application of AI in finance and investment strategies.