Vaibhav Rachalwar

Jr. Data Scientist

Datalink Software

07.2023 - 06.2024

Created 'Sherlock,' an AI platform using Python, Streamlit, OpenAI, and PyMuPDF, enhancing medical coding efficiency by 5-10 times
Utilized OpenCV and Tesseract for OCR to extract text from PDFs/XML, and applied NLP models like LayoutLMv3 and BERT for key information extraction
Implemented parallel processing to optimize performance, showcasing expertise in scalability and resource management also collaborated with a teammate to integrate this with Streamlit interface
Built logistic regression and random forest models to predict Diseases
Used upsampling, hyperparameter tuning with grid search, and ensembling methods (bagging, AdaBoost, gradient boosting) to improve model performance
Created an AI document review system using OpenCV, pytesseract, fuzzywuzzy, and Pandas, including a web interface with Flask, HTML, and JavaScript for navigation and information extraction
Used image processing techniques like contour detection, dilation, and erosion to extract key information from complex layouts
Used NLP techniques like fuzzy string matching to categorize document sections, such as patient records and encounter details
Developed a bookmarking feature for saving document snippets with notes and OCR text, enhancing user experience and collaboration
Developed a Column Mapping Tool using Python, Streamlit, Pandas, and scikit-learn for efficient data integration
Implemented Levenshtein distance, Metaphone phonetic similarity, fuzzywuzzy, and BERT/DeBERTa embeddings to improve mapping accuracy
Created an intuitive Streamlit interface for file uploads, mapping reviews, and manual selections
Developed an interactive healthcare analytics dashboard using Streamlit, Pandas and PySpark, efficiently processing datasets of 100,000+ members and over 1 million rows to providing actionable insights
Leveraged PySpark for initial data processing and complex aggregations, reducing query time for large-scale operations by 70%, while using Pandas for final data shaping and visualization
Employed statistical analysis techniques like Z-score and correlation for hypothesis testing, utilizing PySpark's MLlib for computations on full datasets to identify key performance indicators and trends
Integrated OpenAI's GPT-4 for generating accurate summaries and insights, and optimized code performance using Streamlit's @st.cache_data decorator and PySpark's distributed computing capabilities, reducing overall report generation time by 60%
Worked with stakeholders to develop quarterly roadmaps based on impact, effort and test coordinations.
Utilized advanced querying, visualization and analytics tools to analyze and process complex data sets.