Dhruv Choudhary

Applied advanced data preprocessing techniques to cleanse and enrich scraped job data, ensuring high quality and reliability for subsequent analysis and modeling endeavors. Utilized N-grams methodology to extract nuanced patterns and insights from textual data, enhancing the depth and accuracy of information derived from job descriptions and resumes. Developed and deployed machine learning models utilizing sklearn’s TfidfVectorizer and NearestNeighbors algorithms to construct feature matrices and calculate distances between words from job descriptions and resume skills. Developed a robust database curation process for resume skills using Beautiful Soup. Engineered custom NER models using Spacy to proficiently detect technical skills, libraries, and frameworks, leveraging a comprehensive dataset of over 4,000 entries, leading to a 40% improvement in skills extraction. Developed a machine learning model utilizing Naive Bayes algorithm to accurately classify over 5,000 job descriptions into 10 different job categories with an average accuracy rate of 85%.

Developed and deployed a homography-based forgery detection system using SURF on Aadhaar card images, identifying tampered regions by matching key points and computing descriptors, resulting in an 80% reduction in false positive rates compared to previous methods. Implemented error level analysis, blockiness analysis, and JPEG artefact analysis techniques, identifying and flagging tampered pixels with over 70% efficiency. Applied thresholding and contour detection to identify forged regions in Aadhaar card images based on the calculated difference between genuine and warped images. Conducted extensive research and experimentation, analyzing over 10 research papers and implementing findings into practical forgery detection solutions. Collaborated effectively with cross-functional teams to integrate forgery detection features into Airtel’s document verification systems, ensuring robust security measures.