Data & machine learning

Build an executable analysis or modelling project and inspect the evidence behind its results.

Define the question and the data

Describe the dataset, the outcome you want and the deliverable you will use. For example: “Predict whether an order will be returned from this CSV. Compare the model with a simple baseline and explain the strongest predictors.” Identify the target column and any identifier columns.

Supply a usable dataset

Provide the data or a supported source and explain the column meanings, missing values and relevant dates. The workflow supports dataset preparation and optional external dataset or model sources. External services and GPU execution require the corresponding available connection and runtime.

Agree on evaluation

Choose a metric that fits the question. A supervised project can reserve a separate holdout before training; the platform uses that evaluation data to check predictions. The report should distinguish training results from independent evaluation and explain the baseline comparison.

Read the report and artifacts

The generated project includes executable work and a report preview. Depending on the brief, outputs can include training and prediction code, saved models, predictions and evaluation reports. Inspect the files that were actually produced and the commands needed to reproduce the result.

Handle a failed run

Read the reported problem before requesting a fix. A missing target, unsuitable column types, unavailable dependencies or exhausted runtime can prevent a valid result. Correct the data or brief, run again and check the new report; a chart alone is not proof of a successful training run.

Refine and reuse

Ask for another feature, model comparison or explanation in the same build conversation. Download the generated project when you need to continue in your own environment. Keep the data assumptions and evaluation setup with the result.

Understand the scope

A data project is a different output from a custom AI agent or a built-in Agent Team. A dedicated quantitative trading strategy workflow remains planned; these guides do not describe it as an available builder type.