Mohsen Soleimani

• Conducted extensive research on transformer models for data imputation, focusing on TabPFN, Chronos, MoraiMoE. • Compared the performance of these models against classical forecasting methods like ETS, ARIMA, and XGBoost. • Developed insights into the strengths of modern models in time series forecasting and imputation, contributing to enhanced predictive accuracy.

Led the university’s flagship health-analytics programme, aggregating clinical data from 87 hospitals into a single 100 TB (and growing) warehouse and delivering AI tools that improved patient care, operational safety and reporting speed. • Nation-scale data backbone: Designed an HL7/FHIR-compliant lakehouse (Hadoop 3 / Spark 3 / Kafka 2) ingesting EHR, LIS, PACS, insurance and government-API feeds; automated nightly “shipment logs” validated 100 % of incoming records and kept data freshness under 24 h. • Risk-hazard intelligence: Trained an LLM-powered recommender on HSE standards, chemical-/electrical-/life-safety manuals and five years of incident reports; system flags high-risk units with 90 % precision and routes actionable guidance to hospital HSE teams—cutting hazard-resolution time 35 %. • Radiology AI assistant: Built a UNet + Vision-Transformer pipeline on 12 M DICOM images and 2 M radiologist reports; auto-drafts diagnostic notes that radiologists sign off in 3 days instead of 14—freeing 8 000 + staff-hours per year. • Insurance & government integration: Orchestrated secure APIs that reconcile claims with national payers in < 6 min, reducing manual claim rework 28 %. • Governance & security: Implemented role-based access, differential privacy and on-prem Kubernetes clusters hardened to ISO 27001; zero security incidents in two years. • Stakeholder impact: Dashboards in Power BI/Grafana give clinicians, administrators and regulators real-time KPIs; adoption > 75 % across 87 hospitals. Stack: Python 3.9 · PySpark · Hadoop · Kafka · Airflow · HL7/FHIR · REST/GraphQL · TensorFlow 2 · PyTorch 1.8 · UNet · Vision Transformers · LLMs · Docker · Kubernetes · PostgreSQL / Greenplum · Power BI · Grafana · GitLab CI/CD

• Big‑data backbone: Built a Hadoop 3/Spark 2 lakehouse streaming 20 GB/day from 6 000+ PLC tags; cut data lag 45 min → <3 min. • AI gauge control: Hybrid fuzzy logic + ResNet‑18 at 50 Hz raised thickness precision 25 % and cut off‑gauge scrap 18 %. • Cooling optimiser: XGBoost set‑point model lowered tensile‑strength deviation 18 % and saved 1.2 GWh/yr energy. • Vision QA: YOLOv5/OpenCV on edge GPUs detects 92 % of surface defects; scrap 1.8 % → 1.1 % (≈$420 k/yr). • Predictive maintenance: LSTM autoencoders give 48‑h early‑warning, lifting uptime 12 % (≈$3 M/yr). • Self‑service BI: Django/React/Grafana dashboards for 70+ users cut scheduling time from hours to 20 min. • DevOps & infra: Migrated to vSphere 6.7 (96 cores, 1 TB RAM), Docker/Kubernetes & GitLab CI/CD; release cycles 14 → 3 days. • Leadership: Hired, coached and aligned cross‑functional teams; secured €2.5 M capex to scale AI roadmap. Stack: Python · PySpark · Hadoop · Spark · Kafka · Airflow · IIoT · OPC UA · ResNet‑18 · YOLOv5 · XGBoost · LSTM · Docker · Kubernetes · Grafana