Looking for a job

Payam Soltanzadeh

@payamsoltanzadeh

Message

Data Scientist at DoctorLine, co-building an AI health assistant now used by 1,000+ patients and 12 physicians.

Germany

Message

At DoctorLine, I co-built an AI assistant that answers patient questions through LLM agents, a FastAPI backend, and a Telegram bot. The AI-supported health service has reached 1,000+ patients and 12 active general and specialist physicians.

At Heidelberg University, I built a PyTorch 3D U-Net to denoise radiotherapy simulations; 96.61% of points passed the clinical gamma test. I also designed an evaluation comparing 10 noise-removal methods across noise levels, and built the Python pipeline for preprocessing, training, and validation.

In another Heidelberg University research project, I built sparse recurrent neural networks in PyTorch to learn the Lorenz-63 chaotic time series; the best setting was about 15.7% more accurate than the published baseline. I also led a team in the Initiative D21 / Deloitte National Competition, placing in Germany's top 10 among 200+ teams with an app and companion gadget for simplifying solar setup and use.

Experience

Work history, roles, and key accomplishments

Current

Co-Founder, AI and Data

Current

DoctorLine

Jan 2024 - Present (2 years 9 months)

Co-built an AI assistant using LLM agents with a FastAPI backend and Telegram bot, reaching over 1,000 patients and 12 physicians. Managed backend and network infrastructure and prepared financial and operational reports.

LLM Agents fastAPI Python Backend Development Infrastructure Financial Reporting

Machine Learning Researcher, MSc Thesis

Heidelberg University

Oct 2025 - Sep 2026 (11 months)

Built a 3D U-Net in PyTorch to denoise fast Monte Carlo simulations for radiotherapy planning, achieving 96.61% clinical gamma test pass rate. Designed evaluation comparing 10 noise-removal methods across full noise range and built the Python pipeline.

PyTorch 3D UNet CUDA Python Deep Learning Model Evaluation Medical Imaging

Machine Learning Researcher, Time-Series Models

Heidelberg University

Jan 2025 - Jan 2026 (1 year)

Built sparse recurrent neural networks in PyTorch to learn chaotic time series (Lorenz-63), achieving about 15.7% higher accuracy than the published baseline.

PyTorch RNN Time Series Modeling Sparse Neural Networks Python

Project Lead

Initiative D21 / Deloitte National Competition

Jan 2022 - Jan 2023 (1 year)

Led a team of IT designers and business managers to address Enpal's challenge of making solar energy accessible to elderly users, delivering a working app and companion gadget. Achieved top 10 in Germany among 200+ teams.

Project Management Leadership Product Development Pitching