top of page
Projects
PDF Chat Assistant | RAG System 🔗
Built an AI-powered PDF Question Answering application using Retrieval-Augmented Generation (RAG), enabling users to interact with documents through natural language.
The application extracts and preprocesses PDF text, generates semantic embeddings using Sentence Transformers, retrieves relevant context through FAISS-based vector search, and produces grounded responses using Google Gemini 2.5 Flash. Developed an interactive Streamlit interface for real-time PDF upload and conversational querying.
Real-Time Object Tracking with Motion and Color Detection🔗
Built a real-time computer vision system using OpenCV and Python to detect and track objects based on color and motion. Implemented HSV-based color filtering, contour analysis, and video output generation. The system identifies and labels moving objects in real-time, saving the processed output as a video.
.png)
HindiGPT: A Transformer-Based English-to-Hindi Translation LLM 🔗
HindiGPT is a Transformer-based model built from scratch to provide accurate, context-aware, fluent, and culturally relevant English-to-Hindi translations. It utilizes the encoder-decoder architecture and advanced attention mechanisms to capture complex linguistic patterns between English and Hindi.
Developed a U-Net-based deep learning model for binary segmentation of porosity defects in steel microstructures. Trained on 2,000 real-world X-ray Computed Tomography (XCT) images with OpenCV & TensorFlow, achieving 85% IoU on validation data. Implemented VGG-16 U-Net for enhanced feature extraction and leveraged Otsu’s thresholding for preprocessing. Currently optimizing the model for improved segmentation accuracy.
.png)
.png)
Utilized innovative machine learning techniques to predict diabetes through a meticulously cleaned dataset comprising diverse medical and demographic information. Implemented SMOTE for outcome balancing and employed comprehensive data visualization for enhanced insights. Evaluated eight machine learning models, with Gradient Boosting identified as the most effective for early detection and management of diabetes.
Developed a machine learning model for breast cancer diagnosis classification where I conducted EDA by employing data visualization techniques to create informative visual representations, aiding in feature selection and model optimization. This project aims to predict whether a breast mass is benign or malignant.
.png)
Developed an NLP-based sentiment analysis model using Logistic Regression to classify text as positive, negative, or neutral, achieving an accuracy of 83.77%. Deployed the model using joblib and visualized key insights through plots and word cloud representations during data exploration and analysis.
Designed and executed an unsupervised learning model to explore Facebook post metrics to identify engagement patterns. Conducted data preprocessing and EDA and applied feature selection. Implemented KMeans clustering with the elbow method for optimal cluster determination. Enhanced the model with PCA for dimensionality reduction, and addressed outliers, achieving a silhouette score of 0.6025.
.png)
.png)
Built and optimized a digit recognition system using Keras and the MNIST dataset, delivering precise classification of handwritten digits. Leveraged deep learning methods to refine model accuracy and performance, achieving an accuracy of 99.33%.
Conducted extensive data analysis utilizing Python and subsequently constructed a Machine Learning model designed to discern an individual's adherence to a 'health-conscious' lifestyle.
.png)
.png)
I developed a predictive model to estimate calorie content in the food items and meals of leading fast food chains. Leveraging robust machine learning models and Python's extensive libraries, I conducted comprehensive data analysis and effectively demonstrated the insights through data visualization.
bottom of page