Data Engineer
Resume:
PRIYADARSHINI VIJJIGIRI
linkedin.com/in/vijjigiri adjbk1@r.postjobfree.com github.com/pvijjigiri
WORK EXPERIENCE
Amazon Web Services, Seattle, WA Jun’19 -Sep’20
Software Development Engineer, Analytics and Insights team (VPC - Data Plane) o Release Analyzer: Designed, developed and built an internal tool that takes in all the metrics of the product NX (within VPC) and trains an LSTM model to predict any abnormality in the metrics after the release. This tool helped to rollback many damaging releases before they affect customers.
o Data Pipeline Infrastructure Automation: Built and automated an infrastructure with ETL system that gets data from remote services and finds trends within the data and publishes the analyzed trends in a dashboard using AWS services like Glue (for ETL), Redshift Cluster and QuickSight (SQL queries for analysis). These visualizations helped many internal teams reduce customer issue solve time by 3 days for each issue.
o Per Commit Performance Testing: Analyzed the commits using numpy and pandas libraries to get the insights about changes made to the product. Designed, developed and implemented a component in a testing bot that continuously runs tests on EC2 NX Software on every change committed to it to detect any performance regression or feature breaks. This reduced the debugging time from 1 week to 1 hour. Analog Devices, Boston, MA Jun’18 -Aug’ 18
Data Scientist, Research and Development
o Developed a deep learning-based model (accuracy 93%) using Recurrent Neural networks on the audio data that can detect agitation in the voice which is a part of next generation wearable device for the police. o Applied several signal processing techniques to preprocess the audio data and leveraged existing model built on similar dataset using transfer learning methods.
o Collaborated with system engineers to help productionizing the developed model on real hardware in different environmental and noise conditions.
Oracle Financial Services Software, Bangalore, India Jul’16 -Jul’ 17 Data Scientist
o Built two predictive models using machine learning and NLP techniques, one for scoring customer engagement and other for predicting loan defaults under an umbrella of business use cases with aim to formulate a strategy to curb the customer attrition for a bank (client). o Improved customer experience by optimizing queries within Flexcube software and performed unit testing to identify and resolve issues. o Collaborated with different teams to develop product updates using PL/SQL and JavaScript to ensure on-time delivery. EDUCATION
INDIANA UNIVERSITY, School of Informatics and Computing, Bloomington, IN Aug’17-May ’19 Master of Science, Data Science
o Coursework: Machine learning, Deep Learning, ML for Signal processing, Statistics, Applied Algorithms, Cloud Computing INDIAN INSTITUTE OF TECHNOLOGY (ISM-Dhanbad), India Bachelor of Technology, Electronics and Communication Engineering Jul’12-May’16 TECHNICAL SKILLS
Programming : Python, C++, Spark, SQL, NoSQL, R, Matlab, Ruby, Bash. ML Techniques : Signal Processing, Clustering, Topic Modelling, Recommendation systems, Hypothesis testing Frameworks / Packages : TensorFlow, PyTorch, Airflow, Flask, Tableau, Apache Spark, Docker, AWS, GCP, S3, Glue, Redshift, Scikit-Learn PROJECTS
TensorFlow Speech Recognition Challenge: Aug’18.
o Classified speech signals with an accuracy of 92%, by building a combined model with CNN and LSTM o Made the model more robust to noise level of 5 dB by adding white noise to the data at different SNR values Automated Essay Grader: Apr’18
o Built LSTM based model - for predicting the grade of essay between 0-60 and achieved a best quadratic weighted kappa error of 0.82 o Utilized Genism and NLTK python packages to pre-process the input essays and for converting words to vectors Nuclei Detection Using CNN Architectures: Mar’18
o Efficiently identified several nuclei in images while simultaneously generating high quality segmentation map that contains masks for all the nuclei by building CNN architectures and achieved avg precision value of 0.228 with the ensemble model of U Net and Link Net. Speech Denoising: Apr’18
o Trained a LSTM model on 1200 noisy speech signals with the aim of approximating Ideal binary mask function using truncated backpropagation and achieved an average speech to noise ratio (SNR) of 15dB on 400 test noisy speech signals.
Contact this candidate