Venkata Chivukula

CoLead Contributor to the first Multi Task Graph Foundation ML Framework and design of the model for GCP fleet wide (including GPU and TPU machines) hardware failure Triage (Collaboration with Google Research) CoLead Contributor, design and development of Graph ML model single task which saved thousands of iterations optimizing tech iteration and repair costs by Millions of dollars. (Collaboration with Google Research) Lead Contributor for Infrastructure, Feature Engineering, and productionalizing latest TPU platforms for installations failures in Datacenter floors. giving the same millions of dollars cost savings and thousands of tech iterations reducing average tech iterations. Lead Contributor and Proposed, Designed, and Developed the industry first Adapter+Router Framework to unify Vector DB for RAG applications all independently with minimal supervision. (Collaboration with Cloud AI Research)

Capacity Optimization and Order Prioritization

Doing open-source contributions and building projects

-> Building and optimizing ML pipelines -> Optimize LLM Inference -> Deploying Serverless APIs -> Recommendation System -> Debugging and optimizing containerized images

Teamwork, communication skills, Leadership, Networking, and Action

• Developed TensorFlow Code for the Indian Sign Language Detection model as a part of the project by the Central Government of India. • The project will impact 18 million people for easy navigation of central government websites for health and other online government services. • I led the team and improved word level accuracy by 3% by statistically visualizing the confusion matrix for each of the 100-150 keywords and identifying that the model is not using contextualized embeddings. • Programmed Image Processing script for pre-processing around 1000 (1-2 minute) videos consisting of sign language for easy extraction of image features by the model.

• Created an ML web app aimed at predicting real estate prices using the company’s custom data. • Preprocessed data by filling in missing values, handling categorical variables, reducing dimensionality, and visualizing heatmaps. • Used cutting-edge tools like Git, MLflow, AWS, PySpark MLlib, Docker, Flask, Linux, and HTML/CSS to make the application end-to-end. • The Regression model achieved 97% accuracy

• Assisted in building Backend APIs using Spring Boot for Neo4j Graph Database aimed to support at least 10 million products manufactured by Dell across the globe. • Written Cypher Query Language code for performing basic CRUD operations on the stored knowledge graphs. • Contributed to writing Unit Test Cases for the APIs to ensure the reliability of the application.

• Formulated new Noun Chunking Regex rules for improving the parser’s performance in identifying wrong chunks by less than 1%. • This resulted in the improved performance of Named Entity Recognition and therefore, significantly reducing the error rate for the ATS of the company. • Analyzed more than 1 million documents for writing rules, and developing the word tagging and word clustering algorithm. • Proud to say that MONSTER (A Randstad Company) has been our client and has employed the parser.

Coursework CSE 575 - Statistical Machine Learning CSE 598 - Statistical Learning Theory CSE 579 - Knowledge Representation and Reasoning CSE 511 - Data Processing at Scale CSE 565 - Software Verification, Validation, and Testing CSE 573 - Semantic Web Mining CSE 598 - Data Intensive Systems for Machine Learning CSE 578 - Data Visualization CSE 598 - Engineering Blockchain Applications

Coursework 6.86x - Machine Learning with Python-From Linear Models to Deep Learning 6.431x - Probability – The Science of Uncertainty and Data (Massachusetts Institute of Technology 18.6501x - Fundamentals of Statistics