This project leverages the Gemini API to extract and process YouTube video transcripts, enabling meaningful Q&A interactions based on video content. It utilizes LangChain's YouTubeLoader to fetch transcripts efficiently. The frontend is developed using Streamlit, providing a user-friendly web-based interface, while the backend is powered by Flask, ensuring smooth data processing and API interactions. Additionally, SQLAlchemy is integrated for database management, allowing structured storage and retrieval of transcripts or related data. The project also incorporates NLTK for natural language processing tasks, enhancing the quality of responses. This combination of technologies makes the system robust, interactive, and efficient in delivering context-aware answers from YouTube videos.
In today’s beauty and fashion industries, personalization has emerged as a vital factor in enhancing customer satisfaction and loyalty. Adoption rates remain low due to the lack of automated, accurate personalization tools. A critical aspect of personalization in beauty and fashion is the classification of face shapes, which can guide the recommendation of suitable hairstyles, sunglasses, makeup styles. However, face shape classification is a complex task due to the wide variations in facial structures, poses, lighting conditions, and background clutter. This project addresses the challenge by developing a robust system for female face shape classification into five distinct categories: Heart, Oblong, Oval, Round, and Square. Leveraging deep learning techniques, particularly Convolutional Neural Networks (CNNs) and Transfer Learning, the project aims to achieve high accuracy and generalization across diverse images. Additionally, to offer lightweight alternatives suitable for real-time applications, geometric methods based on facial landmarks extracted through Mediapipe’s FaceMesh solution are explored. Beyond face shape classification, the system integrates a feature-based recommendation engine for personalized sunglasses and hairstyle suggestions, along with a virtual try-on module that enhances user interaction. By combining deep learning models, geometric analysis, and interactive recommendation systems, this project aspires to bridge the gap between consumer expectations and technological advancements in the field of personalized beauty and fashion solutions.
This study explores the application of machine learning techniques for heart disease prediction using the UCI Heart Disease dataset. The dataset, comprising 920 entries with 16 attributes, underwent extensive preprocessing including handling missing values, outlier treatment, and feature engineering. Multiple classification algorithms such as K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Logistic Regression, Decision Trees, Random Forest, Naive Bayes, Gradient Boosting, and XGBoost were implemented to classify patients into heart disease risk categories. The preprocessing pipeline utilized transformations like one-hot encoding, ordinal encoding, and imputation to ensure optimal data preparation. Models were evaluated using metrics such as accuracy, precision, recall, F1 score, and ROC-AUC to identify the most effective classifier. Confusion matrices and visualizations provided insight into the performance of each approach on both training and testing datasets. Results demonstrated varying performance among the algorithms, with ensemble models showing higher accuracy and robustness. The trained models were saved as pipelines to enable deployment in a Streamlitbased application for real-time predictions. This research highlights the efficacy of machine learning in medical diagnostics, particularly for heart disease, and provides a scalable framework for implementation in clinical decision support systems.
Digital watermarking is a crucial technique for embedding and extracting hidden information in digital media, including medical images. Image authentication plays a critical role in ensuring the integrity and authenticity of digital medical images, which are essential for accurate diagnosis, treatment planning, and research. The project focuses on the development of a robust watermarking algorithm for images authentication using methods, including techniques like Haar Transform, Histogram Shifting, Arnold’s Cat Map. The methods aim to embed an imperceptible watermark into images, which can later be extracted to verify the authenticity and integrity of the images.