Tanmay Chaturvedi

Spearheaded development of an ML-driven transformer fault diagnosis system, achieving 93-95% accuracy and processing 100+ mat files with a customizable preprocessing pipeline.

Formulated a feature extraction framework, computing 5+ features (e.g., Kurtosis, median, impulse factor) on customized datasets, processing up to 1.2 million samples.

Orchestrated FFT and wavelet transform visualizations for 5 wavelet types, enabling real-time adjustments via Streamlit.

Devised a Supervised Contrastive Learning pipeline for feature extraction and SVM classifiers, supporting 5 activation functions across 1000+ epochs.

Contributed to an Advanced Retrieval-Augmented Generation (RAG) pipeline, attaining 92% efficiency across 500+ queries while enhancing document processing capabilities.

Executed over 500 alpha-phase test cases to validate robustness, preparing for beta testing of 1,500+ queries.

Integrated PaddleOCR, realizing an 85% success rate in data extraction for the RAG pipeline.

Co-authored a research paper detailing methodologies and performance enhancements of the RAG pipeline project.

Evaluated machine learning models to detect Non-Performing Loans (NPLs), identifying the Random Forest Classifier with the highest success rate on a dataset of over 250,000 users.

Benchmarked 7 classifiers to enhance credit risk assessment frameworks and improve predictive accuracy.

Authored and published a paper, "Comparative Performance Analysis of Machine Learning Algorithms for Non-Performing Loan Prediction," accepted at IEEE ICCCIT conference.

Awarded the Best Research Paper Award at IEEE ICCCIT conference for outstanding contributions.