Medical imaging
Read the volume.
Prepare volumetric cardiac MRI, train a 3D segmentation model, and report validation Dice with its dataset limits.
Explore cardiac MRI segmentation ↗Medical AI · Statistical programming · Humanities & psychology
Medical AI Statistics for Humanities
I'm Seyyed Arad Hosseini Moghaddam, an AI engineer focused on machine learning for medicine, especially 3D medical imaging. I also apply statistical programming and data science to questions in humanities fields such as psychology. Explore documented projects, methods, and reproducible results.
About
Medical imaging meets quantitative human inquiry.
My primary work applies deep learning to medical imaging, including 3D cardiac MRI segmentation, training, and rigorous benchmark evaluation.
I also focus on statistical analysis and Python-based data science for psychology and other humanities questions. My wider engineering work spans Persian NLP, LLMs, RAG, agentic AI, and reproducible deployment. Public artifacts document methods, evaluation, and limitations.
Research pipeline · conceptual illustration
From complex signals to reviewable evidence.
Three connected practices shape my work. The diagrams below illustrate the workflow; they are not patient images or model outputs.
Capabilities
Medical AI, statistical data science, and engineering
01 / Learning systems
Deep Learning and Medical AI
- PyTorch
- TensorFlow
- Computer Vision
- Medical Imaging
- CNNs
- Segmentation
- Model Evaluation
- Python
- SQL
- Statistics
- Data Science
- Humanities Data Analysis
- NLP
02 / Language models
Generative AI and Fine-tuning
- LLMs
- Transformers
- Hugging Face
- Fine-tuning
- LoRA
- QLoRA
- PEFT
- Prompt Engineering
- Embeddings
- RAG
- Vector Databases
- Semantic Search
- LangChain
03 / Orchestration
Agentic AI and LLM Orchestration
- Agentic AI
- AI Agents
- LangGraph
- LangSmith
- Deep Agents
- ReAct Agents
- Tool Calling
- Multi-Agent Systems
- Subgraphs
- Stateful Agents
- Conversational Memory
- Human-in-the-Loop
- LLM Observability
- LLM Cost Optimization
- MCP
04 / Delivery
MLOps
- Git
- FastAPI
- MLflow
- Docker
- GitHub Actions
- CI/CD
05 / Product engineering
Software and Web Development
- HTML
- CSS
- JavaScript
- Tailwind CSS
- Django
- PyQt6
- Flutter
06 / Quantitative research
Statistics and Humanities Data
- Applied Statistics
- Statistical Analysis
- Exploratory Data Analysis
- Humanities Data Analysis
- Quantitative Research
- Data Cleaning
- Data Interpretation
Selected work
Research, with evidence.
Explore cardiac MRI segmentation, Persian language-model training, and a multi-agent research platform, with methods, metrics, and clear limitations.
01 / Flagship model · Persian NLP · From-scratch training · PyTorch · Transformers
Persian GPT — 110M Transformer From Scratch
A 110.4M-parameter Persian Transformer trained from scratch, with a custom tokenizer, baselines, and held-out evaluation.
110.4Mparameters
244Mtraining tokens
14.52held-out perplexity
Perplexity evaluates language modeling, not factual accuracy.
02 / Medical AI · 3D MRI · Deep learning · PyTorch · MONAI
3D Cardiac MRI Segmentation
3D cardiac MRI segmentation with PyTorch and MONAI, from volumetric preprocessing to 3D U-Net training and benchmark evaluation.
0.8821best validation Dice
100training epochs
3D U-Netarchitecture
Benchmark validation only; not clinically validated.
03 / Agentic AI · RAG · Evaluated system · LangGraph · LangChain
AI Research Assistant Platform
LangGraph agents for retrieval, cited research synthesis, and tool use, with FastAPI endpoints and reproducible offline tests.
103offline tests
5agent roles
REST + SSEAPI surface
Offline tests verify software behavior, not answer quality.
Research directions
Medical machine learning and humanities data
Deep learning for medical applications
3D MRI segmentation, transparent validation, and reproducible medical machine-learning workflows.
View cardiac MRI work ↗Statistics and data science for humanities
Python, statistics, and exploratory analysis for psychology and humanities datasets. An active research direction, not a claimed case study.
Read research directions ↗FAQ
A few useful details.
What kinds of AI systems have you built?
I have built agentic AI and RAG systems, including tool-using workflows, retrieval pipelines, vector search, stateful interactions, and LLM orchestration. My work also includes traditional machine learning and deep learning.
Do you train deep learning models from scratch?
Yes. I have worked across computer vision, NLP, and medical imaging. One example is PersianGPT, a 110.4M-parameter decoder-only transformer trained for Persian.
What is your strongest computer vision experience?
My focus includes medical imaging, segmentation, and 3D pipelines, with work across CNNs, preprocessing, inference, visualization, and task-specific evaluation.
Do you fine-tune language models?
Yes. I have worked with full fine-tuning as well as LoRA, QLoRA, and PEFT. Models and related work are available on Hugging Face.
What NLP and language model work have you done?
My experience includes language model training from scratch, tokenizer and dataset preparation, transformer fine-tuning, structured extraction, embeddings, semantic search, RAG, and LLM agents.
Do you develop mobile applications?
Yes. I develop Android and iOS applications with Flutter.
Where can I find your source code?
My public repositories are available on GitHub.
Where can I find your models and datasets?
My public models, checkpoints, and datasets are available on Hugging Face.
Which projects best represent your work?
Persian GPT demonstrates end-to-end language-model training; 3D Cardiac MRI Segmentation demonstrates medical-imaging deep learning; and the AI Research Assistant demonstrates evaluated multi-agent RAG engineering. Their public pages document methods, results, and limitations.
Contact
Let’s build something intelligent.
Have a project, research idea, or opportunity in mind? Send me a message.