Sai Tulasi Kolapudi
Sai Tulasi Kolapudi
📍 Denton, Texas
✉️ Email: saitulasi1729@gmail.com
🎓 Education
-
University of North Texas | Denton, USA
Master’s in Data Science (CGPA: 3.90)
Aug 2022 - May 2024 -
Saveetha University | Chennai, India
Bachelor’s in Computer Science (CGPA: 3.45)
Aug 2016 - Dec 2021
🔍 Skills
- Languages: Python, Java, SQL, NoSQL, R
- Developer Tools: Google Collab, Jupyter Notebook, Azure DevOps (ADO), MLflow
- AI/ML Frameworks: PyTorch, TensorFlow, LangChain, LangGraph, LangSmith, RAGAS, Hugging Face Transformers, PEFT, LORA
- LLM Technologies: RAG, Agentic AI, LLM Evaluation, Fine-tuning, RLHF, Prompt Engineering, Vector Databases
- Cloud & Deployment: AWS (SageMaker, Bedrock, Lambda), Azure ML, GCP, Docker, Kubernetes
- Big Data: Apache Spark, Hadoop, ETL Pipelines, Data Preprocessing
- Visualization & BI: Matplotlib, Seaborn, Tableau, Power BI
- NLP/Computer Vision: BERT, GPT, YOLOR, OpenCV, Named Entity Recognition
💼 Experience
- JLL | Data Scientist - P2 | Dallas-Fort Worth Metroplex
June 2024 - Present- Built a comprehensive evaluation framework from the ground up for assessing AI-powered information retrieval systems
- Developed platform-wide evaluation capabilities enabling consistent quality measurement across all intelligent solutions
- Created cost estimation models to optimize AI deployment efficiency and forecast resource utilization
- Revolutionized the Signals Company Matching system by developing and implementing delta matching algorithms that reduced fuzzy matching computational overhead by 98%
- Led analysis of existing data correlation implementations and developed roadmap for next-generation capabilities
- Currently leading development of agentic capabilities for JLL GPT, enhancing autonomous reasoning and task execution
- Capital One | Machine Learning Engineer Intern | Remote
March 2024 - May 2024- Developed and implemented Kai Bot within a Slack group of 20,000 users, significantly automating the assignment of technical incidents, reducing manual ticket handling by over 70%
- Implemented a Retrieval-Augmented Generation (RAG) pipeline to enhance the accuracy and relevance of automated responses by combining generative and retrieval-based AI techniques, increasing solution accuracy by 30%
- Integrated GPT-based AI solutions to automate responses and handle frequent inquiries, improving engagement and problem resolution rates within the Slack group
- Implemented embedding models for categorizing and suggesting pre-existing solutions, reducing the average incident closure time by 50% through better initial response accuracy
- Priviom | Machine Learning Engineer Intern | Remote
December 2023 - March 2024- Developed and implemented advanced data preprocessing techniques and utilized vector databases, significantly enhancing data efficiency and model accuracy for SaaS cost optimization
- Fine-tuned Large Language Models (LLMs) with state-of-the-art methods like delta-tuning and LoRA, achieving predictive insights into SaaS usage patterns to drive cost reductions
- Integrated AI models into AWS cloud environments, optimizing scalability and performance, leading to a marked improvement in the cost-efficiency of SaaS applications
- University of North Texas | Research Associate | Denton, USA
Aug 2022 - May 2024- Spearheaded the development of PatentGPT, a novel patent summarization model, by fine-tuning MixTRL 8x7b through RLHF and PEFT, leading to a significant 40% accuracy improvement over existing open-source solutions
- Pioneered the implementation of Quantization and Low-Rank Adaptation (LORA) techniques on PatentGPT, successfully reducing model size by 80%, optimizing resource utilization for large-scale deployment scenarios
- Fostered international collaborations by presenting original research on advanced patent Summarization using large language models at the prestigious IEEE QRS conference in Thailand, 2023
- Enhanced patent summarization using advanced NLP models (ChatGPT, Llama) for accurate and efficient results
- Extended context length of Language Models (ChatGPT, Llama) for improved text generation quality
- Explored Federated Learning for privacy-preserving model deployment on edge devices
- Constructed automated data pipelines in Python for real-time model training and evaluation
- University of North Texas | Graduate Student Assistant | Denton, USA
Aug 2022 - May 2024- Supported the INFO 5501: Fundamentals of Data Science course as a Graduate Student Assistant
- Led tutorial sessions and lab exercises in Python programming, data visualization, and machine learning
- Developed and graded assignments, quizzes, and provided constructive feedback to students
- Collaborated with instructors to improve curriculum, integrating real-world examples and projects
- Gained experience in pedagogy, contributing to an engaging educational environment
- DXC Technologies | Data Engineer and Machine Learning Engineer | Chennai, India
Jun 2020 - Aug 2022- Accomplished a 50% reduction in model deployment time and enhanced deliverable acceleration by implementing scalable ML models on AWS Sagemaker, using TensorFlow and PyTorch, which led to improved scalability and reliability
- Achieved a 25% improvement in predictive model accuracy, resulting in an estimated annual savings of $550,000 in operational costs, by refining deep learning algorithms and neural network applications
- Led to a 60% increase in data processing efficiency and significant reduction in database operational costs by optimizing data pipelines using Apache Spark and Hadoop
- Ensured robustness and accuracy in AI applications by designing and executing A/B testing frameworks for model validation, which led to continuous improvement in model performance in real-world scenarios
- Fostered a culture of innovation and technical excellence within the team by curating and delivering workshops on machine learning and cloud-native AI technologies, which enhanced team capabilities and fostered knowledge sharing
- Designed and implemented cloud-based machine learning solutions using AWS Sagemaker and Python, ensuring scalable and efficient model deployment
- Developed robust ETL pipelines in Python to synchronize data across SQL Server, Oracle, and Postgres databases, streamlining data availability for analytics
- Conducted hyperparameter tuning and feature selection, leading to an improvement in model performance metrics by over 20%
- Applied natural language processing (NLP) techniques for sentiment analysis and text classification using a variety of machine learning algorithms, enhancing data interpretation efforts
- Leveraged cloud computing (AWS), Python programming, database management systems (DBMS), natural language processing (NLP), and machine learning skills to drive innovative solutions
🚀 Projects
-
Lane Detection using YOLOR and OpenCV July 2023 Real-time lane detection system developed in Python using YOLOR and OpenCV. The project enables users to select a Region of Interest (ROI) manually for targeted lane identification. It employs Canny Edge Detection and Hough Line Transform to accurately identify and track lanes in video streams. Suitable for real-world applications like Advanced Driver-Assistance Systems (ADAS).
-
Patent Document Summarization Jan 2023 Developed neural network models for summarizing patent documents based on research insights. Explored models include T5 variations, XLNet, BART, BigBird, Pegasus, and GPT-3.5. Evaluated using various metrics like ROUGE, BLEU, and BERT Score. Implemented using Hugging Face Transformers and Seq2Seq.
-
Breast Cancer Prediction Aug 2022 Predicting breast cancer diagnosis using various machine learning techniques. Achieved impressive accuracy rates with different models including RobustScaler Naive Bayes, PCA Test, and more.
-
Secure_Ecommerce Sep 2023 The Grocery Store Web Application is a Flask-based web application that allows users to browse, search for, and purchase grocery products online. It also provides features such as user registration, authentication, wishlists, and order history tracking. It is designed to provide a comprehensive demonstration of Flask capabilities in conjunction with MongoDB, PayPal integration, and email functionalities.
-
Document-Processor-Web-Application Apr 2024 The Document Processor is a web application designed to extract and process information from call logs. Users can input a question and upload call log documents, from which the application will extract relevant facts. It offers features like time navigation, auto-approval of facts, and handling contradictions in the extracted information.
📚 Publications
- Deep Belief Network and Sentimental Analysis for Stock Market
- Quality Evaluation of Summarization Models for Patent Documents
- Sentiment Exploration System to Improve Teaching and Learning
🌟 Extracurricular
- UNT Data Science Organization | Secretary | Denton, USA
Sep 2022 - May 2024