Nervousness Detection System:
Objective: To classify facial images into confident and unconfident categories using a Convolutional Neural Network (CNN).
Challenges: Handling variations in facial expressions, lighting conditions, and image quality to accurately detect nervousness.
Tools/Techniques: Utilized CNNs, image preprocessing, Label Encoder, and Haarcascade for face detection.
Outcomes: Achieved a training accuracy of 95.49% and validation accuracy of 76.22%, demonstrating the model's potential in real-time nervousness detection.
Musafir_bot (AI-powered travel chatbot):
Objective: Developed an AI-powered travel chatbot for the `Hack2Skill` hackathon to provide instant, personalized travel guidance.
Challenges: Ensuring accurate travel recommendations, handling diverse user queries, and integrating AI-powered language processing.
Tools/Techniques: ReactJS, API from aimlapi.com, and modern web technologies for a seamless experience.
Outcomes: A real-time, user-friendly chatbot offering comprehensive travel information, rapid responses, and intelligent recommendations for exploring India.
Shail-Shri (Booking and Information Website of Uttarakhand):
Objective: Develop SHAIL-SHRI, an information and booking portal for Uttarakhand's tourist destinations, as capstone project for O-Level (IT).
Challenges: Managing diverse information for multiple destinations.
Tools/Techniques: Web development tools (HTML, CSS, JavaScript).
Outcomes: Streamlined booking process for visitors. Contribution to tourism and ease of travel in Uttarakhand.
Analyze Death Age Difference of Right Handers with Left Handers:
Objective: Analyze and compare the age of death between right-handed and left-handed individuals to identify potential patterns or differences.
Challenges: Overcoming potential biases in data collection, considering cultural factors, and addressing the complexity of defining handedness accurately.
Tools/Techniques: Statistical analysis using tools like Python, survey data collection methods, and demographic analysis.
Outcomes: Uncover insights into whether there is a significant difference in the age of death between right-handed and left-handed individuals, providing valuable information for further research or public health considerations.
Web Scraper ( DOJ Document Analysis Pipeline ):
Objective: To develop an automated system that extracts, processes, and analyzes legal documents from the U.S. Department of Justice archives using Natural Language Processing and machine learning techniques for efficient insight generation and metadata extraction.
Challenges: Managing and processing large volumes of legal PDFs with varying formats, quality issues, and complex legal terminology while maintaining system stability and accuracy in a concurrent processing environment.
Tools/Techniques: Utilized BART for text summarization, SpaCy for NLP, PyPDF2 for document extraction, aiohttp for async downloads, PostgreSQL with SQLAlchemy for data management, and asyncio for concurrent processing implementation.
Outcomes: Achieved 92.7% accuracy in document classification, 99.3% processing completion rate, 87% summary relevance score, and 94.5% metadata extraction accuracy while reducing document processing time by 78% through concurrent processing optimization.
AI-Powered Writing Assistant:
Objective: To create an intelligent writing assistant that leverages the GPT-2 language model to generate creative writing content across multiple genres with customizable parameters based on user prompts.
Challenges: Balancing text generation creativity with coherence, managing model response time for real-time interactions, handling diverse writing styles across different genres, and ensuring consistent output quality regardless of prompt complexity.
Tools/Techniques: Utilized GPT-2 medium model for text generation, PyTorch for deep learning framework, Transformers library for model implementation, and Gradio for creating an intuitive user interface with real-time text generation capabilities.
Outcomes: Successfully implemented a writing assistant supporting five distinct genres (Sci-Fi, Fantasy, Historical Fiction, Mystery, and Romance) with customizable text length, achieving coherent narrative generation with adjustable creativity parameters through temperature and top-k/p sampling controls.
Book Recommendation System:
Objective: Develop a Book Recommender System using k-Nearest Neighbors and Popularity-Based models as a MCA capstone project. Guide users to books based on prolific user ratings and popular titles.
Challenges: Managing diverse book datasets. Fine-tuning algorithms for relevance. Balancing recommendations from different models.
Tools/Techniques: Python, scikit-learn, pandas, numpy, k-Nearest Neighbors algorithm.
Outcomes: Enhanced user experience with personalized book recommendations. Potential business opportunities for delivering tailored suggestions. Successful integration of collaborative and popularity-based models.