Projects

CNN Feature Visualizer

Built a Python library and containerized React web app for a cloud-based CNN Feature Visualizer, enabling developers to visualize and analyze layers of CNN models for improved training and insights. Utilized Docker, Kubernetes, Redis, Minio, PostgreSQL, and Flask to design a scalable and distributed architecture for the CNN Feature Visualizer, facilitating efficient processing and storage of model data and visualizations. Deployed the CNN Feature Visualizer using Google Kubernetes Engine (GKE) cluster.

CafeConnect

Designed and built CafeConnect, a full-stack student restaurant review app with a notification system, utilizing Spring Boot and Java for the backend, React and Python for the front end, and Cockroach DB for data storage. Utilized software design patterns like MVC, Singleton, Strategy, and Observer to address complexities in the project, highlighting adaptability and problem-solving skills in software design.

Daily News Summarizer

Developed a personalized daily news summarizer utilizing innovative technologies such as FastAPI, Langchain, Pinecone, and OpenAI API. Utilized Pinecone for vector database storage and implemented a Retrieval-Augmented Generation (RAG) pipeline for content generation. Implemented a user-centric approach allowing for customization based on individual preferences, including semantic search functionality and article classification. Enhanced user experience by delivering tailored news summaries and key points, optimizing information consumption.

Imdb Chatbot

Developed an IMDb chatbot using Python, Langchain, Streamlit, and GPT 3.5, offering users a conversational interface to retrieve information about movies and TV shows from the IMDb database, using a Retrieval-Augmented Generation (RAG) pipeline. Implemented natural language processing capabilities to understand user queries and extract relevant details efficiently. Enhanced user experience through interactive features and concise text summarization of movie information.

Arcane

Text-based RPG game featuring diverse user-controlled adventurers, various creatures to combat, a range of treasures to collect, and four distinct floors to explore. Implemented key Software design patterns, including Observer, Decorator, Singleton, Factory, and Strategy, to enhance the game's architecture and functionality.

music-seperation-service

This repository contains the implementation of a Kubernetes cluster providing a REST API for an automatic music separation service called Music-Separation-as-a-service (MSaaS). The provided services include a Flask backend, REST frontend, worker nodes for music analysis, a Redis database for task queuing and storage using Min.io.

Text-2-SQL-LLM

Fine-tuned Llama-2 and Llama-3 models for a text-to-SQL task using LoRA and QLoRA, significantly improving model accuracy from 31.8% to 60.6%. Implemented advanced post-processing techniques to enhance SQL query generation accuracy, surpassing the baseline performance of GPT-3.5 and GPT-4 models.