Agentic AI Research Assistant
The Agentic AI Research Assistant is a personal project inspired by a common challenge in research: efficiently finding, organizing, and interacting with growing collections of information. I developed this project to explore modern agentic AI concepts by combining large language models with retrieval-augmented generation (RAG) and modular tool integration. The current version enables users to query a local document collection through a conversational interface, retrieving relevant information before generating context-aware responses. Beyond building a functional application, this project allowed me to gain hands-on experience with agentic workflows, prompt engineering, and modular AI system design. It also serves as a foundation for future enhancements, including web search, scientific literature retrieval, multi-agent collaboration, and integration with external knowledge sources.
Technologies & Methods: Python • Large Language Models (LLMs) • LangGraph • OpenAI API • Retrieval-Augmented Generation (RAG) • Agentic AI • Prompt Engineering • Tool Calling • Semantic Search • Modular AI Architecture • Git & GitHub
GitHub: https://github.com/denizalacam/agentic-ai-research-assistant
EHR Readmission Prediction with Machine Learning and Deep Learning
This project demonstrates the development of an end-to-end predictive analytics pipeline for 30-day hospital readmission prediction using synthetic electronic health record (EHR) data. Starting with raw patient, encounter, diagnosis, medication, procedure, and observation data, I built a complete workflow encompassing data preprocessing, feature engineering, model development, and performance evaluation. The project compares classical machine learning models, including Logistic Regression and XGBoost, with a deep learning model developed in PyTorch, providing insights into the strengths and limitations of different predictive approaches. Beyond model development, the project highlights practical experience in healthcare data analytics, predictive modeling, explainable machine learning, deep learning, and the development of reproducible AI workflows for real-world clinical applications.
Technologies & Methods: Python • Pandas • Scikit-learn • XGBoost • PyTorch • Predictive Modeling • Machine Learning • Deep Learning • Feature Engineering • Model Evaluation • Explainable AI • Healthcare Data Analytics
GitHub: https://github.com/denizalacam/ehr-readmission-prediction
Digital Twin Modeling for Personalized Radiotherapy
This project focuses on developing patient-specific digital twin models to support personalized radiotherapy and precision medicine. By integrating mathematical modeling with quantitative medical imaging—including CT and PET—my work aims to simulate tumor behavior, predict treatment response, and explore individualized treatment strategies. The project combines computational modeling, predictive analytics, parameter optimization, and simulation to translate complex biological processes into reproducible digital representations of individual patients. Through model development, validation, and cross-validation, this work demonstrates how mathematical modeling, medical imaging, and artificial intelligence can be integrated to advance data-driven clinical decision-making and improve personalized cancer care.
Technologies & Methods: Python • MATLAB • Mathematical Modeling • Digital Twins • Medical Imaging • Personalization • Precision Medicine • CT Imaging • PET Imaging • Medical Imaging • Predictive Modeling • Simulation • Parameter Optimization • Cross-Validation • Scientific Computing • Data Analysis • AI
Representation Learning for Functional Brain Connectivity
Understanding how different regions of the brain communicate is essential for advancing neuroscience and improving the diagnosis of neurological and psychiatric disorders. In this project, I developed a novel computational framework for analyzing dynamic brain connectivity from resting-state fMRI data. The approach combines representation learning through Independent Component Analysis (ICA), signal processing, spectral analysis, and statistical modeling to characterize time-varying interactions between functional brain networks. Applied to a large multi-site neuroimaging dataset, the framework identified distinct connectivity patterns associated with schizophrenia, demonstrating how computational modeling and machine learning can uncover clinically relevant biomarkers and provide new insights into brain function.
Technologies & Methods: MATLAB • Python • Resting-State fMRI • Neuroimaging • Representation Learning (ICA) • Functional Connectivity Analysis • Biomarker Analysis • Signal Processing • Spectral Analysis • Dynamic Brain Connectivity • Statistical Modeling • Machine Learning • Data Visualization • AI
Publication: https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2023.1078995/full
Environmental Systems Modeling & Simulation
Understanding and predicting the behavior of complex environmental systems requires integrating experimental research with mathematical modeling. In this work, we developed a compartmental mathematical model based on differential equations to investigate the effects of copper-induced stress on biological systems and evaluate their potential for environmental remediation. By combining experimental observations with computational modeling, parameter estimation, optimization, and sensitivity analysis, we developed a predictive framework to simulate contaminant accumulation, biological responses, and long-term system dynamics. The resulting model provided quantitative insights into environmental processes and demonstrated how mathematical modeling and simulation can support environmental research and sustainable remediation strategies.
Technologies & Methods: MATLAB • Mathematical Modeling • Differential Equations • Compartmental Modeling • Dynamical Systems • Predictive Modeling • Simulation • Parameter Estimation • Optimization • Sensitivity Analysis
Publication: https://www.sciencedirect.com/science/article/abs/pii/S0304380022000849
Modeling Rhythm Generation Mechanism in Small Neural Circuits
Central pattern generators (CPGs) are specialized neural circuits that autonomously generate rhythmic motor behaviors such as swimming, walking, breathing, and heartbeat. Understanding how these circuits produce robust and adaptable behaviors requires computational models that accurately capture their underlying biological mechanisms. In this work, we developed a highly detailed, biologically plausible computational model of the swim CPG in the sea slug Melibe leonina, creating a realistic digital twin that reproduced experimentally observed neural activity and rhythmic behaviors. Building upon this foundation, we developed simplified computational neural network models to investigate the fundamental principles of rhythm generation, multistability, robustness, and bifurcation dynamics. While the detailed model faithfully replicated the biological system, the simplified models enabled systematic exploration of the rich repertoire of rhythmic behaviors that neural circuits can exhibit, providing broader insights into the dynamical principles governing biological neural systems.
Technologies & Methods: MATLAB • Python • C • Electrophysiology • Digital Twins • Conductance-Based Neural Modeling • Computational Neuroscience • Differential Equations • Dynamical Systems • Bifurcation Analysis • Neural Network Simulation • Scientific Computing • GPU Computing • Mathematical Modeling • Model Validation
GitHub: https://github.com/denizalacam/Plant-CPG-Model
Publications: https://www.worldscientific.com/doi/abs/10.1142/S0218127415400039
https://pubmed.ncbi.nlm.nih.gov/32752614/