Vurukonda Sathish

Developed an Isolation Forest model to detect anomalies in manufacturing data which systematically recorded from various sensors during different phases of each product’s production cycle. Developed a table detection algorithm capable of identifying tables within machine or scanned documents and extracting the text.

Developed a document extraction product that leverages large language models (LLMs) like GPT and PALM and achieved an impressive extraction accuracy rate of over 92%. Developed an efficient inference engines for RNN, LSTM, GRU, and BERT model in C++ with object-oriented programming, CuBLAS, and CuDNN library and achieved an inference speed of over 70% as compared to traditional Python inference libraries. Involved in field extraction project from structured and unstructured documents which is based on deep learning models and transformers and achieved an impressive accuracy rate of 90%. Built an ensemble model with Logistic Regression, Random Forest, XGBoost, Catboost, SVM, and ANN models on the extracted fields from documents to provide a conversion probability for each new submission with an impressive accuracy rate of over 93% and built a regression model to prioritize new submissions. Optimized the field extraction solution for new submissions and achieved a 70% reduction in inference time compared to the original implementation and 40% reduction in their size without any loss of accuracy after applying quantization techniques. Developed an efficient inference engine for extracting fields from new account submissions using inference engines RNN, LSTM, GRU, and BERT model and achieved a remarkable speed improvement of 75% compared to the PyTorch implementation. Developed an ensemble model with Logistic Regression, Random Forest, XGBoost, Catboost, SVM, and ANN models multi-model for predicting the ratings of company pros and cons based on Glassdoor reviews and achieved an impressive accuracy rate of over 93%. Designed an ensemble of classic machine learning models to predict the confidence scores for all recognized fields on bank checks and model achieved an impressive 65% confidence score for every field on the check.

Assisted in designing and/or conducting tutorials, assignments, experiments and grading the quizzes for: Estimation and Identification (EE638), Non-linear Systems (EE613), Oprimal Control (EE622), Control Systems Lab (EE324).

Research centered around the modeling of multivariate time series data with a particular emphasis on count data.