Data Science Machine Learning

OM SAI KRISHNA MADHAV LELLA

517-***-**** ******************.*****@*****.*** LinkedIn GitHub Google Scholar EDUCATION

Michigan State University Master of Science in Data Science Aug 2023 - May 2025

(Courses: Data Mining, Big Data, Statistical Modeling, Optimization, Computer Vision, Natural Language Processing) CGPA: 4.0/4.0 Indian Institute of Technology Madras Bachelor of Technology in Chemical Engineering Jul 2015 - May 2019

(Courses: Algorithms, Linear Algebra, Probability and Statistics, Time Series Analysis, Multivariate Data Analysis) CGPA: 3.41/4.0 TECHNICAL SKILLS

Programming Languages : C, C++, Java, Python, R, Matlab, React.js, Redux, JavaScript, TypeScript, HTML/CSS Python Packages : NumPy, SciPy, Pandas, Scikit-learn, Matplotlib, PyTorch, TensorFlow, OpenCV, NLTK, Transformers Frameworks & Tools : LangChain, LlamaIndex, Kafka, Spark, Selenium, ETL, CUDA, RESTful APIs, Springboot, IntelliJ, Git Deep Learning : FNN, CNN, RNN, LSTM, Transformers (BERT, GPT, ViT), GAN, Autoencoders, Stable Diffusion Databases : PostgreSQL, MySQL, MongoDB, GraphQL, Oracle DB, SQLite3, Cassandra, Hive, Elasticsearch Cloud Technologies : AWS (SageMaker, EC2, EMR, RDS, S3), GCP (Pub/Sub, BigQuery, Composer), Azure (Databricks) PROFESSIONAL WORK EXPERIENCE

AI/ML Research Assistant Michigan State University, University of Illinois Urbana-Champaign Feb 2024 - Present

• Analyzed transcriptions from ~250K YouTube videos by leveraging sentence transformer embeddings and accelerating dimensionality reduction and clustering with GPU-enabled UMAP and HDBSCAN algorithms to discover latent topics.

• Utilized quantized LLMs for few-shot classification and fine-tuned LLMs to effectively detect stance expressed in tweets.

• Developed a Chrome plugin to visualize browsing data and deployed a Django-based Passive Data Kit on EC2 for data collection. Technologies: Python, JavaScript, Django, PostgreSQL, AWS EC2, AWS RDS, CUDA, Transformers, LLMs, LLM Quantization Junior Data Scientist II Gyan Data Mar 2022 - Jun 2023

• Implemented XGBoost with feature engineering to predict equipment faults, reducing maintenance-related downtime by 50%.

• Integrated Hugging Face Transformers to replace Word2Vec/GloVe models for text processing tasks, boosting performance by 15%.

• Performed LLM-assisted content analysis on a million text documents to contextualize and optimize search within local systems.

• Outperformed BERT-based models, boosting the F1 score from 0.90 to 0.95 in multi-class text classification with a linear SVM.

• Deployed JupyterHub on an Ubuntu server, configuring dependencies, SSL, authentication, and systemd services for scalable multi- user access to Jupyter Notebooks, while integrating machine learning models with IPywidgets for interactive UI access for all users.

• Developed a usage analytics dashboard to track usage statistics and monitor the health of machine learning models in real-time. Technologies: Python, JupyterHub, Docker, Apache Server, Nginx, SSL, Linux, CI/CD, IPywidgets, XGBoost, SVM, Transformers Technology Analyst Citi Aug 2019 - Mar 2022

• Developed a Content Composition Tool for analysts to streamline creation and publishing of financial articles on CitiVelocity.

• Implemented a hybrid recommendation system to suggest articles to CitiVelocity users, increasing engagement by 25%.

• Used K-means, DBSCAN, and Autoencoders to detect abnormal trading behaviors, achieving up to 0.95 in all performance metrics.

• Developed a RAG-based LLM agent to answer queries and raise tickets for human support when issues remain unresolved.

• Engineered a scalable data pipeline to process billions of records daily, using Apache Spark for distributed processing, Hive for storage, cron jobs for scheduling, and real-time monitoring to ensure system reliability and performance.

• Improved the performance of data transformation jobs by 40% by coalescing small files before writing them to the data warehouse. Technologies: Python, Java, BERT, React.js, TypeScript, HTML, CSS, Springboot, REST APIs, Postman, Kafka, Spark, Hive, CronJob Star Intern Wipro May 2018 - Jul 2018

• Developed and optimized ETL workflows in Talend to preprocess securities transaction data and detect wash trades.

• Implemented algorithms to parse ETL job XML files and map property details to migrate ETL jobs from DataStage to Talend. Technologies: Java, React.js, Springboot, Maven, SQLite3, HTML, CSS, JavaScript, MySQL, ETL, Talend, DataStage, Hadoop, XML PROJECTS

Speaker Diarization Sparta Hack 9 Track Winner

Engineered an automated speaker diarization system for Zoom calls by integrating computer vision and natural language processing to track speakers in real-time, synchronize audio with identified speakers, and generate user-specific transcriptions and summaries. Graph-Based Retrieval-Augmented Generation for Question Answering GitHub Repository Developed a Knowledge Graph-based RAG system to enhance LLM performance by constructing, visualizing, and querying interconnected knowledge graphs from text documents, thereby improving entity search accuracy and context comprehension. Momentum-Based and Adaptive Optimization Methods in Deep Learning GitHub Repository, Project Presentation Video Implemented and performed a comparative analysis of optimization methods, including SGD, Adam, and others, in deep learning. 3D Scene Reconstruction for Autonomous Robot Navigation Best Project, Computer Vision Course GitHub Repository Developed a robust 3D navigation system for autonomous robots, integrating depth estimation using MiDaS, 3D scene reconstruction with RGB-D data, real-time object detection with YOLOv8 Nano, instance segmentation with Mobile SAM, and optimized path planning using the A* algorithm, ensuring efficient and obstacle-free navigation in dynamic environments.

Contact this candidate