Coastal Typhoon Inundation Surrogate Model using LangChain and Vast.ai Cloud GPU
With El Niño altering Pacific sea surface temperature gradients, severe typhoons produce unpredictable storm surges along coastal cities in East Asia. Standard hydrodynamic numerical solvers (e.g., Delft3D, ADCIRC) take hours to execute per storm track scenario, preventing real-time insurance claims loss reserving and evacuation planning. In this challenge, you will construct a high-speed machine learning surrogate model pipeline orchestrated by LangChain and trained/deployed on high-performance GPU nodes via Vast.ai. The workflow will map pressure minimums, radius of maximum winds, and bathymetry grids directly to coastal flood inundation maps in milliseconds. Your solution will execute physics-informed neural network (PINN) inferences, estimate insured property losses, and explain flooded zone predictions via an LLM agent interface.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a fast flood surrogate model pipeline with LangChain deployed on Vast.ai GPUs to predict typhoon storm surge coastal damage.
How work is evaluated
Evaluates inundation depth model speedup and depth prediction RMSE against hydrodynamic solver reference outputs.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
latency_check
Surrogate model execution completes in under 2 seconds
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
surge_depth_rmse
Root Mean Squared Error of predicted coastal flood depth • target: 0.25 • range: 0-5
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Build a LangChain multi-step chain for surrogate model inference and analysis.
Provision and deploy deep learning compute pipelines on Vast.ai.
Train a spatial neural network surrogate for hydrodynamic flood simulations.
Evaluate physical property loss against spatial flood depth maps.
Reference links and supporting material
Synthetic grid dataset of coastal elevation profiles, bathymetry raster maps, and hydrodynamically simulated flood outputs across major East Asian ports.
Challenge setup
Evaluates inundation depth model speedup and depth prediction RMSE against hydrodynamic solver reference outputs.
Challenge input
JSON with storm_central_pressure_hpa, wind_speed_kts, landfall_lat_lon
LangChain
Mandatory framework for chain and model workflow orchestration.
Vast.ai
Provides scalable, low-cost cloud GPUs for surrogate model execution.
Evaluated output
JSON with max_surge_depth_m, inundated_area_sqkm, estimated_loss_usd
- Surrogate model execution completes in under 2 seconds
- Root Mean Squared Error of predicted coastal flood depth • target: 0.25 • range: 0-5
- Benchmark: SurgeSimBench
- Surge Depth Rmse target: 0.25
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Vast.ai · Required
- Langchain · Optional
- Anthropic · Optional
- Langchain · Optional
Evaluation contract
- latency_check · Weight 1
- surge_depth_rmse · Weight 1
Metrics
- surge_depth_rmse · Target 0.25
Benchmarks
- SurgeSimBench
- materials
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with storm_central_pressure_hpa, wind_speed_kts, landfall_lat_lon
Output: JSON with max_surge_depth_m, inundated_area_sqkm, estimated_loss_usd
Execution harness: Python
Evaluation tools
- LangChain · Agent Orchestration Framework
- Vast.ai · Inference Acceleration Host
Recipe freeze marker
No recipe freeze time is recorded.