I'm Sepehr Rezaee, an AI researcher, engineer, and innovator passionate about building robust, secure, and impactful machine learning systems.

• 🏫 BSc in Computer Science, Shahid Beheshti University (2021–2025)
• 🧠 Research Intern @ Mackenzie W. Mathis Lab, EPFL (2025–Present)
• 🤖 AI Engineer @ Agentic Systems / PropTy Global (2024–Present)
• 🔬 Former Research Assistant @ Sharif University of Technology & Shahid Beheshti University

My focus areas:

• Deep Learning & Computer Vision: Generative modeling, diffusion models, robust AI, adversarial defense, and AI safety.
• Physics-Informed Neural Networks (PINNs): Applying deep learning to scientific and biomedical challenges.
• Large-Scale, Multi-Agent LLM Systems: Designing scalable, production-grade AI architectures.
• Secure and Trustworthy AI: Backdoor/trojan detection, model interpretability, and reliable ML in critical systems.

See more in my CV.

• Research Intern @ Mackenzie W. Mathis Lab (EPFL, 2025–Present)

  • Co-authored an ICCV 2025 accepted paper, introducing DISTIL: a data-free, diffusion-driven framework for trigger inversion in Trojaned neural networks—setting new SOTA on BackdoorBench (+7.1% acc) and object detection scanning (+9.4%).
  • Developed novel, safe, and interpretable generative modeling pipelines.
  • Pioneered zero-shot, data-free defenses for backdoor attacks, advancing reliable machine learning for mission-critical use.
  • Contributed to empirical evaluation and benchmarking for trustworthy AI.

• AI Engineer @ Agentic Systems / PropTy Global (2024–Present)

  • Architected and deployed multi-agent LLM platforms using LangChain, RAG, FastAPI, Docker, and Kubernetes.
  • Achieved 85%+ task completion rates in autonomous business decision workflows.
  • Reduced onboarding time by 15%, improved contextual relevance, and delivered sub-100ms real-time API performance.

• Research Assistant @ Robust & Interpretable Machine Learning Lab, Sharif University of Technology (2024–2025)

  • Authored and submitted 3 papers to NeurIPS 2024 on ML security and reliability.
  • Developed robust pipelines for adversarially resistant ML, including in autonomous driving and healthcare.
  • Presented at international conferences, expanding academic collaboration.

• Research Assistant @ AI & Scientific Computing Lab, Shahid Beheshti University (2023–2025)

  • Published and under-review papers on disease modeling, Fokker-Planck equations, and Alzheimer’s detection with MRI.
  • Used PINNs for integrating physical laws into neural networks.

• Deep Learning & Neuroscience Intern @ IPM (2023–2024)

  • Led M/EEG signal decoding with advanced deep learning.
  • Improved neural network architectures and research workflows.

• DISTIL: Data-Free Inversion of Suspicious Trojan Inputs via Latent Diffusion (ICCV 2025, accepted)
• Scanning Trojaned Models Using Out-of-Distribution Samples (NeurIPS 2024, accepted)
• Comparison of Pre-Training and Classification Models for Early Detection of Alzheimer’s Disease Using MRI (I4C 2023, accepted)
• Multiple under-review and conference papers on robust optimization, data-free backdoor detection, and scientific ML.

📰 See my Google Scholar and CV for the full list.

• Head Teaching Assistant: Advanced Programming, Data Mining & Analysis (Shahid Beheshti University)
• Teaching Assistant: Basic Programming, project mentor for AI in industry (petrochemical, water, electricity)

• Best Ideator Award: 7th National Young Scientists Festival (2023) – AI-based early Alzheimer’s assistant
• Top 0.2% National Entrance Exam: Placed 352nd out of ~150,000 (2020)

• Email: [email protected]
• GitHub: SepehrRezaee
• LinkedIn: linkedin.com/in/sepehr-rezaee/
• Personal Website: sepehrrezaee.com