Predictive Mining Capital Allocation
With battery prices hitting record lows and commodity prices like gold and oil fluctuating, mining companies must decide which capital-intensive bulk projects to prioritize. This challenge tasks you with building a decision-support system that evaluates mining project viability. You will use the Mistral Nemo model for advanced multilingual analysis and TorchServe for deploying a custom price-prediction model. The core of the application will be a vector database that stores complex project profiles, including capital expenditure (Capex), geographical risks, and mineral types. You will implement a RAG pipeline that allows users to query which 'Megamines' are most resilient to a low-battery-price environment. Additionally, you will explore the use of the NNI (Neural Network Intelligence) toolkit to optimize the hyperparameters of a regression model predicting mineral price trends for 2026.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a mining investment evaluation engine using Mistral Nemo, Weaviate, and NNI to analyze capital-intensive projects against shifting commodity price benchmarks.
How work is evaluated
The evaluation checks the integration between Weaviate and Mistral Nemo, as well as the optimization performance of the NNI tuning process.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Design a Weaviate schema to store mining project attributes such as 'Capital Intensity', 'Jurisdiction', and 'Mineral Category'.
Implement a RAG pipeline using Mistral Nemo to interpret Goldman Sachs' 2026 commodity outlook in the context of specific zinc projects.
Optimize a PyTorch regression model using NNI's 'Evolutionary Algorithm' tuner to predict the Internal Rate of Return (IRR) for new mines.
Orchestrate model serving by containerizing Mistral Nemo and the regression model using TorchServe.
Build a custom Weaviate Module to automate the vectorization of project feasibility studies.
Implement advanced filtering logic to compare project viability across different battery metal price scenarios (e.g., $50/kWh vs $100/kWh).
Utilize the NNI 'Trial Management' dashboard to track the performance of different model architectures for price prediction.
Create a Gradio front-end that allows executives to adjust 'Trade Dependency' weights and see real-time impacts on project rankings.
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
The evaluation checks the integration between Weaviate and Mistral Nemo, as well as the optimization performance of the NNI tuning process.
Challenge input
String query + Price scenario parameters
Agent execution
No agent tool is listed for this challenge.
Evaluated output
Ranked list of mining projects with justification
- System must successfully query the vector store and return relevant project metadata.
- Verification that NNI ran at least 10 trials for hyperparameter optimization.
- Mean Absolute Error (MAE) target: 50
- Python execution harness
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
No tool records are attached.
Evaluation contract
- System must successfully query the vector store and return relevant project metadata.
- Verification that NNI ran at least 10 trials for hyperparameter optimization.
Metrics
- Mean Absolute Error (MAE) ยท Target 50
Workflow tasks
- Task 1
Input: String query + Price scenario parameters
Output: Ranked list of mining projects with justification
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.