Medical AI · machine learning · quantitative data science
An evidence-led archive of my work in deep learning for medical imaging, statistics and data science for humanities-related questions, Persian language-model training, and broader AI engineering. Featured projects include reported metrics, methods, and limitations where the source project documents them.
Research profile · graduate study · open portfolio
I focus on deep learning and machine learning for medical applications, particularly medical imaging. I also pursue statistics and data science for humanities-related datasets. These directions and the public projects below do not imply an academic affiliation, publication, or clinical validation.
01 / Medical machine learning
Medical imaging and 3D segmentation are the primary focus. The work emphasizes data preparation, model architecture, validation, inference, and transparent reporting of benchmark results and their limits.
Cardiac MRI segmentation project ↗02 / Quantitative humanities
A research direction applying statistical analysis and data-science workflows to humanities-related data. The portfolio currently describes this as an area of interest rather than presenting an unverified completed case study.
Related methods and skills ↗Selected evidence
These projects show different parts of my technical range. Metrics are reproduced from the public project records and should be interpreted within each project's dataset and evaluation setup.
Flagship model · Persian NLP · From-scratch training
This project demonstrates an end-to-end low-resource language-model workflow for Persian, from corpus preparation through tokenizer design, training, and held-out evaluation.
The reported perplexity is specific to the held-out corpus and tokenizer. It is not a measure of factuality, instruction following, or downstream task performance.
Medical AI · 3D MRI · Deep learning
The project investigates voxel-level cardiac structure segmentation from volumetric MRI, a setting where models must use three-dimensional spatial context.
The reported validation Dice is from this project’s benchmark split. The system has not been clinically validated and is not a clinical decision-support tool.
Agentic AI · RAG · Evaluated system
This system explores how retrieval, web research, tools, synthesis, and bounded reflection can be orchestrated and tested as a stateful research workflow.
The offline suite validates application contracts with deterministic model behavior. It does not establish comparative research-answer quality or clinical safety.
Research approach
I aim to make technical work reviewable: state the problem, describe the data and model, report the evaluation protocol, publish reproducible code or artifacts, and make limitations visible. This portfolio supports academic review, prospective graduate study, and research collaboration.
Loading projects
Project type
Research area