Fix PhysicsNemo PINN Flow Field Model (3D geometry) - Output Violates Physics

We have a working Physics-Informed Neural Network (PINN) setup using NVIDIA's [PhysicsNemo](https://docs.nvidia.com/deeplearning/physicsnemo/getting-started/index.html) framework. The simulation runs correctly, but the **output violates physical laws** — velocity in a narrowing throat region decreases, which contradicts continuity.

---

### What we have:

- 3D STL geometry and mesh
- Physics-informed training code using `physicsnemo`
- Loss setup already includes Navier-Stokes terms
- Trained model outputs, but unphysical flow field
- Directory includes: `fengdong_flow.py`, `geometry.py`, `properties.py`, `conf/`, STL, outputs
- Archive file (275MB): [DOWNLOAD LINK HERE - INSERT YOUR LINK]

---

### Your Task:

1. Inspect the code (esp. loss and boundary conditions)
2. Fix the setup so that flow follows expected physics:
- e.g., velocity increases in narrow section (mass conservation)
3. Retrain or partially retrain the model
4. Plot flow field slice results and loss curves
5. Provide Python scripts for post-processing

---

### Deliverables:

- Corrected training code
- Training outputs that reflect realistic fluid behavior
- Plots of velocity field and training loss (e.g. matplotlib)

---

### Skills Needed:

- PhysicsNemo or PINN experience
- PyTorch
- Fluid dynamics / Navier-Stokes equations
- Docker (optional)

---

### Deadline:

3 days max.

---

Please include any relevant PINN projects you've done in your bid. We are looking for someone who understands both ML and physics.