JAGATH KATAMA

EDUCATION

Indiana University, Luddy School of Informatics, Computing, and Engineering Indianapolis, IN
- Master of Science in Applied Data Science
- Expected Graduation: Dec 2025
National Institute of Technology, Rourkela, Odisha, India
- Bachelor of Technology in Electronics and Communication Engineering
- Graduation: May 2023

Currently working on confidential medical project under the guidance of Prof. Hee - Tae Jung in the Health Laboratory Lab involving advanced NLP and Generative AI techniques, with responsibilities spanning from model development to production deployment.

Provided feedback to Applied Data Science Master’s students on statistical analysis and R programming assignments, enhancing their conceptual and practical skills.

Worked as a TA for INFO-I 308: Information Representation (Undergraduate Course). where I instructed undergraduate students in database design principles and SQL using MySQL and phpMyAdmin. Evaluated student assignments and provided technical support.

Developed Python code generation models using lightweight large language models (LLMs), including GPT-2, LLaMA 3.2, Gemma 3, and Qwen 2.5. Fine-tuned over 200K+ Python code samples using Parameter-Efficient Fine-Tuning (PEFT) with LoRA and quantization techniques on multi-GPU setups (NVIDIA A16) using the Accelerate library.
Designed and executed experiments to benchmark model architectures (124M–3B params), optimizing training via gradient checkpointing and evaluating outputs with cross-entropy loss and code-specific tasks.
Identified performance tradeoffs across model families and exposed limitations in metrics like perplexity and CodeBLEU, contributing to research on efficient, local LLM-based code generation.

Designed and developed a Transportation Management Database for managing bookings, vehicle and driver assignments, invoicing, and SaaS subscriptions for a road-based transport company.
Implemented role-based access controls and business rules to ensure data security and system integrity, including constraints for vehicle and customer assignments.
Optimized database structure through normalization (1NF to 3NF) to reduce redundancy and enhance data integrity.
Created detailed documentation outlining database schema, relationships, business rules, and SQL setup instructions for seamless integration and use.

Designed a content-based filtering model to analyze user preferences, generating personalized apartment recommendations (from nearly 10k apartments).
Applied Principal Component Analysis (PCA) and pairwise similarity metrics, improving the efficiency of recommendation generation by 20% while reducing computational overhead.
Implemented a count-vectorizer to improve search capabilities, reducing search query processing time to 0.5 seconds,10% compared to initial baseline models.
Deployed the model as an interactive web application using Streamlit, with visualizations to enhance user engagement.

Formulated testable hypotheses (ANOVA, chi-square tests, t-tests) to analyze the relationship between BMI and lifestyle factors (more than 12 factors).
Conducted regression analyses on multiple response variable types (continuous, count, proportion/rate) to evaluate the impact of lifestyle variables on obesity risk.
Interpreted statistical findings for reliability and validity to derive actionable insights.

Programming languages: Python, R, SQL (Advanced), C/C++, Bash Scripting
Developer Tools: Tensorflow, VSCode, Pycharm, Pyspark, Tableao, Excel, Docker, Rstudio
Cloud Platform: AWS
Big Data Engineering: Apache Spark, Pyspark, Apache Kafka
ML/AI Framework: Tensorflow, Pytorch, HuggingFace, Transformers
Database: MySQL, SQLite, MongoDB, NoSQL, VectorDB
Libraries: Pandas, Numpy, PyTorch, keras, tensorflow, Scikit-learn,Asyncio, Hugging Face, Seaborn, requests, BeautifulSoup, Selenium, transformers, Datasets, NLTK, SpaCy, Accelerator, Langchain