My Projects
AI-Based Sketch to Webpage Converter
Description: Built an end-to-end DL pipeline to convert UI sketches to HTML using YOLOv8 for object detection, custom CNN for UI-aware features, Transformer with self-attention for hierarchy modeling, and GRU for tag generation. Integrated EasyOCR for text extraction and achieved mAP@50 = 0.96 with support for theme customization and layout hierarchy.
Technologies: Python, PyTorch, YOLO, CV, NLP, SketchIT Dataset
Link: GitHub
AI-Powered GitHub Code Review Bot
Description: Built a backend service with FastAPI and GitHub Webhooks to automate PR reviews, integrating API validation and reducing manual review effort by 40%.
Technologies: Python, FastAPI, GitHub, OpenAI API
Link: GitHub
Abstractive Summarization of Legal and News Documents
Description: Fine-tuned a BART model on BillSum and Multi-News datasets to generate concise summaries. Achieved ROUGE-1: 47%, BLEU: 16%, and BERTScore: 88% through hyperparameter tuning and efficient training.
Technologies: Python, PyTorch, LLM, NLP
Link: GitHub
Building an Audio Classification Model with STFT and TensorFlow
Description: Explored how to develop an audio classification model using Short-Time Fourier Transform (STFT) and TensorFlow, covering data preprocessing, spectrogram generation, and deep learning techniques for sound recognition
Technologies: Python, Tensorflow, Librosa, ESC-50 Dataset
Temporal Trends in Hospital Admissions - The Influence of Air Quality on Respiratory Health
Description: Conducted a data-driven analysis to investigate the relationship between air pollution (PM2.5, PM10, NO2, SO2) and hospital admissions for respiratory illnesses using five machine learning models. Logistic Regression model helped detecting admissions linked to air quality issues occurring six months prior. Findings provided actionable insights for improving healthcare system preparedness and resource allocation
Technologies: Python, scikit-learn, Pandas, Numpy
Link: GitHub
Enhanced Pedestrian Detection
Description: A convolutional neural network model to enhance pedestrian detection in occluded and hazy weather conditions.
Technologies: Pytorch, Python
Steganography Messenger Mobile Application
Description: Engineered an Android mobile application that allows users to securely send encrypted messages by embedding them into RGB images leveraging AES Algorithm.
Technologies:Android App Development, Android Studio, Security Principles
SQL Parser
Description: Created a SQL parser model capable of performing syntax analysis on input SQL statements using LEX programming. This tool streamlines SQL error detection and debugging by providing efficient syntactic validation, improving database query management for developers.
Technologies:LEX Programming