Turn consumption time series into useful forecasts.

The model supports univariate and multivariate forecasting with covariates. You can study history and compare horizons within its permitted uses. The linked official weights are non-commercial and non-production; using results in operations or commercial decisions requires Google’s authorisation.

A configuration supporting covariates can combine production history with weather available for the forecast horizon. Check which variables are actually known at that point. Evaluate each plant on separate data and retain the source of weather forecasts used.

Hold out periods following the training or context history and compare by horizon, site and operating conditions. Use a baseline such as the last value or a seasonal pattern. Review errors around peaks and missing data alongside the overall average.

The base model supplies image features; it does not include a detector for your site’s defects. You can use those representations in an adapted system with your examples. The task and capture conditions determine the training and evaluation needed before integration.

Retain the asset identifier in the incident and document metadata. Your application retrieves the relevant manual and procedure, and an LLM can prepare a referenced summary. Results connect to the maintenance system your team manages.

It depends on whether you query documents, analyse images or process signals with a specialist model. Estimate frequency, volume and concurrency for each stage. Start with a set of assets and compare total workflow cost with the value of its information.

Your integration can use the system’s APIs or exports to provide data and receive results. Preserve asset identifiers and timestamps. QDivZero runs compatible models; your application links outputs to maintenance records and workflows.

They serve different tasks. An LLM can help query manuals, summarise incidents or explain available information. Sensor predictions need a model suited to those signals and a specific evaluation. You can connect both components in an application while keeping their functions distinct.

Choose a recurring task with accessible data and a verifiable result, such as querying asset manuals or classifying work records. Define which decision the output supports and who reviews it. This scope lets you evaluate usefulness and integration before adding sites or models.