Seyed Hasan Mir Hosseini
Data Scientist / Machine Learning Engineer
Professional Summary
Data Scientist with 2+ years of hands-on experience building and deploying predictive and deep-learning models (Transformers, CNN, LSTM) in Python. Experienced with large time-series datasets, feature engineering, and production deployment using Docker, CI/CD, and Azure ML across the full ML lifecycle.
Core Skills
| Languages | Python (primary), SQL, R |
| LLM / Agents | LangChain, LangGraph, RAG systems, Multi-agent orchestration, Prompt engineering, Groq (LLaMA 3.1), Gemini, ChromaDB, sentence-transformers, BM25 + semantic hybrid retrieval |
| MLOps / Deploy | Docker (multi-stage), Docker Compose, GitHub Actions CI/CD, Azure ML managed online endpoints, Azure App Service, REST APIs, Git |
| ML / DL | PyTorch, Transformers, CNN, LSTM, Autoencoders, Self-supervised learning (VICReg), scikit-learn, LightGBM, XGBoost, CatBoost |
| Data | Pandas, NumPy, SciPy, Matplotlib, Feature engineering, Time-series processing |
| Web / Backend | Django, REST APIs, HTML, CSS |
Work Experience
- Built a Transformer-based foundation model on large time-series data, pretrained with self-supervised learning (VICReg) and fine-tuned for classification.
- Developed the full PyTorch pipeline: preprocessing, augmentation, pretraining, fine-tuning, and evaluation.
- Designed and compared predictive deep-learning models (LSTM–Transformer, CNN–Transformer, and multimodal fusion) for time-series classification.
- Implemented end-to-end pipeline in PyTorch: preprocessing, spectrogram & time-series feature extraction, training, and model evaluation.
- Processed and analysed large experimental datasets: cleaning, preprocessing, statistical analysis, and visualisation.
- Co-authored peer-reviewed publications based on the research findings.
Selected Projects
Personal portfolio website with live ML model prediction demos, a RAG-powered AI assistant, and a technical blog. Deployed predictive ML models to production on Azure App Service with live prediction endpoints. Built a modular inference registry routing requests to Azure ML endpoints, with automatic local-model fallback.
Longitudinal biomarker trend analysis over time-series medical records. RAG assistant with hybrid BM25 + semantic retrieval and multi-LLM failover (Groq, Gemini).
Multi-agent RAG pipeline orchestrating 6+ specialised agents for multilingual content generation, integrated with a Django web interface. Built during a hackathon.
Competition-grade tabular and time-series prediction. Gradient-boosting ensembles (LightGBM, CatBoost, XGBoost), Random Forest, and Autoencoders. Feature engineering and cross-validated optimisation; evaluated with ROC-AUC, RMSE, and F1.