Develop AI surrogate models that learn from high-fidelity CFD outputs to deliver near real-time predictions of flow metrics for vertiport planning (downwash footprint, peak velocities, recirculation zones, building/ground effects, safety envelopes).

You will:

• Build a clean, versioned dataset pipeline from CFD outputs + configuration metadata (geometry class, rotor params, layout, wind BCs, turbulence settings).
• Define inputs/outputs aligned with vertiport decisions (spacing/layout, envelopes, weather sensitivity).
• Train baseline surrogates and improve via architecture/conditioning/losses (e.g., spatially weighted, physics-inspired regularization).
• Assess generalization to new layouts/weather/drone configs and add uncertainty awareness (ensembles/calibrated confidence).
• Package a lightweight inference component that integrates with the platform and CFD validation workflow.

Required skills:

• Must-have: Strong PyTorch; solid ML engineering (clean loops, logging, checkpoints, reproducibility); experience with spatial/temporal data; scientific data formats (HDF5/NetCDF/VTK-like); strong evaluation and error analysis.
• Nice-to-have: Neural operators/GNNs; uncertainty quantification; CUDA/inference optimization; familiarity with CFD concepts.

Deliverables:

• Dataset + data contract, loaders, and documentation; versioned train/val/test splits without leakage.
• Trained surrogate models + benchmarks (baseline + improved), latency/memory profiling.
• Final report and reproducible repo (training/eval/inference/configs), safe-use domain and validation summary.

Suggested plan (6 months):

• M1: Data contract + dataset pipeline v1 + targets/metrics + split strategy.
• M2: Baseline surrogate(s) + first E2E results + error analysis.
• M3–M4: Model improvements + robustness/generalization.
• M5: Expand scenarios + uncertainty estimation + reliability analysis.
• M6: Consolidation + validation + final report + handover (+ publication-ready figures if applicable).

Compensation: Monthly stipend, with potential performance bonus and paper co-authorship.

📧 Pour postuler: contact@redxt.com