Vantipuli Aishwarya

Cincinnati Children's Medical Research Center

• Developed an automated bone age interpretation tool that predicts bone age on more than 20,000 trauma hand radiographs and achieved MAE of ±6 months using NVIDIA MONAI deep learning framework.
• Streamlined Multi-GPU training pipelines to build different experiments for finding the optimal network architecture and hyperparameters using data versioning tools like DVC, Weights and Biases & Pachyderm.
• Published and presented a poster on A Microservices based open-source MLOPs Stack for Medical Imaging Research at conference SIIM 2022-Society for Imaging Informatics in Medicine.
• Tech stack - Python, DVC, MONAI, Tableau.

SAP America, Newton Square, PA

• Developed time series ARIMA models for analyzing and forecasting sales of several consumer products at large scale retail organizations.
• Designed and developed solutions by understanding the manufacturing industry workflow and acted as the subject matter expert by recommending suggestions to the automation of the tasks performed at the warehouse plant.
• Extracted meaningful business insights from data and identify the stories behind the patterns, distilling complex analysis, and concepts into concise business-focused takeaways using Tableau.
• Researched new technologies and methods across data science, ML, and data engineering to improve technical capabilities such as detecting/correcting invalid physical addresses.
• Tech stack - Python, MySQL, SAP HANA, Timeseries, Tableau.

Northeastern University, Boston

• Developed the Work pad, a device type and orientation agnostic charging pad using Deep Learning in Embedded Systems.
• Generated on-board devices’ heatmaps using LSTM+CNN Time-Series model given input signal distributions; eliminated noisy frequencies using Custom Low Pass Filters and Fast Fourier Transforms.
• Researched and implemented efforts in data generation using GANs, VAEs to extend model to new devices and positions and reduce data collection time by 65%

Younify, Hyderabad, India

• Proposed Business insights for NISSAN using Instagram Analytics. Data is extracted using Instagram API
• Developed Image Recognition Application for Image organization and classification of photo libraries using Computer Vision and Keras which helped in attracting and retaining customers.
• Designed an Interactive chatbot using Google Dialog Flow. Increased bot’s performance by 60% by integrating various APIs and Machine Learning to give best user experience.

Northeastern Univeristy

GPA: 3.6/4

Coursework:
Supervised Machine Learning
Unsupervised Machine Learning
Deep Learning
Natural Language Processing
Probability and Statistics
Algorithms ans Datastructures
Database Management Systems
Python for Data Science

Bone age Prediction

• Accomplished preprocessing Normal CCHMC full training dataset (Predicting Chronological age from trauma hands) Investigated multiple cleaning techniques to remove the lettering artifacts, enhancing contrast of X-rays by implementing CLAHE.
• Utilized Multi GPU parallelization for training and implemented it for boneage workflow. Some of the profiling tools used are Deep Learning Profiler (DLProf)​, NVIDIA Nsight Systems.
• Generated Low resolution images by capturing primitive shapes and colors in Stage-I GAN. In Stage-II, High Resolution images are produced by training a GAN on low res images. Accelerated training performance using Horovod, a deep learning multi gpu framework on boneage project and also built Open MPI for High performance computing.
• Streamlined boneage workflow in new open-source framework by MONAI. Created a pipeline to load, preprocess, train, evaluate the data by designing a single training script which works for different architectures and parameters.
• Expanded the model optimization capability and integrated boneage workflow by using new tool Weights and Biases (Wandb) for experiment tracking.
• Minimized the Mean Absolute Error on RSNA test data by 54% (from 11 to 5 months) using MONAI framework. Implemented unit testing.
• Bolstered AI-ARC team on technical issues and error solving methods related to wandb, cuda, python and git.

Text to Image Generation

• Developed a two-stage generative adversarial network for generating photo realistic images from textual descriptions on Caltech-UCSD Birds (CUB) Dataset by training Generator & Discriminator in a Min-Max game.
• Converted text to vector representation by learning word embeddings using Reccurent Neural Network.
• Generated Low resolution images by capturing primitive shapes and colors in Stage-I GAN. In Stage-II, High Resolution images are produced by training a GAN on low res images.
• Measured performance of the model using Inception score, KL divergence loss along with human evaluation. Honored with best capstone project in class award among 20 groups.

Plagiarism Detection

• Examined 100K documents which contains textual answers given by students which were labeled with different levels of plagiarism
• Pre - processed and Clustered similar words using K-means clustering. Created word-embeddings using GloVe for vector representation of words
• Implemented FP-Growth Algorithm to get frequent item sets from transactions in the form of vectors.
• Built a Recurrent Neural Network and predicted short summaries of speeches using End-End Memory Network and Position Encoding methods with an accuracy of 91.4%

Diabetic Retinopathy Detection

• Created an automatic DR grading system capable of classifying images based on disease pathologies from four severity levels using Image Classification.
• Pre-processed Images using OpenCV, Otsu’s Method by removing Gaussian blur, boundary effects and cropped to isolate the subject. Normalized images to represent pixels between 0 to 1.
• Compared performances of Logistic Regression, Convolutional Neural Network (CNN), KNN and Multi-layer Perceptron (MLP) and Google’s Inception v3 on High Resolution Retina Images.
• Achieved an accuracy of 93% with CNN. Confusion Matrix, F1 score, ROC and AUC metrics are used to evaluate the model

Employee Review Analysis

• Analyzed 10 years of Employee Review Data scrapped from Glass Door, containing textual and Numeric reviews by current and former employees of Top 6 companies.
• Handled missing data using Mean Value Imputation and SMOTE on training data to handle class imbalances
• Generated geo-spatial data for the employee locations using Google ggmap Package and represented them on an interactive World Map using Leaflet Package.
• Developed a Regression model to suggest companies about the areas they can improve based on employee perspective by reporting high and less correlated features.

Mini Datascience Projects Using R

• Conducted Sentiment Analysis on Trumps Speeches. Observed Trends along different timeframes and visualized the data using R packages.
• Investigated U.S. Transgender Data and understood interesting insights like causes for suicides and homelessness within different categories like gender, race etc.
• Visualized presence of different types of radio active materials in American Indian reservoirs in the southwestern United States and determined their risk levels
• Analysed bibliographic information on major computer science journals from DBLP database. Implemented MYSQL in R using RMYSQL and RSQLite packages to pull the data from database.

Northeastern University

Tutor and instruct students on concepts in Algorithms and Python for Data Science and coordinate with Professor in creating, grading the assignments and exams.

Women in Cyber Security (WiCS)

As a volunteer initially and then later becoming the head of the Technical Wing at Northeastern University (Computer Science Dept), I have organized several Intra and Intercollege coding competitions and also conducted other technical events, quiz and event planning. I also managed the volunteers assigned to this group to plan the events frequently.

Council for Green Revolution, India

During my association with the organization, I have contributed as a volunteer, co-coordinator. I have participated and raised concerns about environmental isuues. I have actively engaged in technical summer camps and Sunday schools that provide a common platform for students, industrialists, and educationalists to interact and exchange knowledge about the trending technological advancements in agrculture.