I bridge the gap between advanced Computational Fluid Dynamics (CFD) and Scientific Machine Learning. My primary focus is developing hardware-agnostic surrogate models (Neural Operators, PINNs) optimized for Model Predictive Control (MPC) and energy transition applications.

Driven by reproducible science, I engineer pure-Python inference pipelines that transform state-of-the-art literature into lightweight, vectorized solvers capable of executing live directly in the browser via Pyodide.

• CFD Researcher @ BSC (Barcelona Supercomputing Center)
• M.Sc. Fluid Mechanics (Grenoble INP)
• Inventor (US Patent 11,124,953: Thermal/Fluid Systems)

To overcome the "technical debt" often found in academic codebases, my repositories strictly adhere to a decoupled architecture. I separate the core mathematical engines from project-specific deliverables, ensuring scalable, production-grade research software:

Saeid-Bahrami/ (Profile Anchor & Reproducibility Hub)
│
├── ⚙️ SciML-Thermal-Core/            # THE ENGINE: Single Source of Truth
│   ├── sciml_core.solvers/           # Vectorized NumPy baseline physics solvers
│   ├── sciml_core.neural_ops/        # Barebone architectures (FNO, DeepONet)
│   ├── sciml_core.mpc_controller/    # Model Predictive Control (MPC) optimization loop
│   └── sciml_core.pyodide_bridge/    # WASM serialization for Web UI rendering
│
├── 📂 Archetypes/       # THE DELIVERABLES: Micro-Projects & Live Deployments
│   │
│   ├──  Project_1_HVAC_Surrogate/    # Archetype: Neural surrogate for building energy MPC
│   │   ├── weights/                  # Pre-trained lightweight checkpoints (.npy / .pt)
│   │   ├── main_inference.py         # Lightweight script calling SciML-Thermal-Core
│   │   └── index.html                # <py-script> integration for LIVE browser execution
│   │
│   └──  Project_2_Digital_Twin/      # Archetype: Sensor fusion & data assimilation framework
│       └── ...
│
└── ...

Hardware-Agnostic Inference: Heavy training is decoupled. The archetype repositories rely on pre-trained weights and dense NumPy vectorization to perform lightning-fast inference, proving architectural viability without requiring HPC access.

Zero-Install Reproducibility: By bridging Python algorithms with WebAssembly (Pyodide), physical simulations and neural network inferences can be evaluated instantly by researchers and reviewers globally, directly in their browsers.