CO2 emissions prediction projects can help central energy managers and production site owners better inform energy usage and production strategies, but interpretation of resulting analytics is time consuming and often limited to expert data teams. Now with Generative AI — and more specifically in this use case, large language models (LLMs) — these teams can use self-service exploration to instantly query data outcomes and answer questions more specific to their own facility or domain.
- Empower Regional Teams: Different regions can ask simple questions in the language of their choice with outputs in that language.
- Gain Efficiency: Improve processes by reducing the time and complexity required to develop reports and insights.
- Reduce Impact: Uncover trends and problematic sites faster to increase response time.
- Better Planning: Improve production planning at local and global levels with machine learning-based predictions.
How It Works: Architecture
A project predicting CO2 emissions runs in Dataiku. The energy management team and any relevant personnel can then interact with the output of the prediction project through a chatbot interface, which generates visuals based on the queries. Examples include questions like:
- Show me the overview of electricity consumption.
- Show me the electricity consumption breakdown by site.
- What is the scenario that minimizes the CO2 emissions?
- Comparer la production des sites de Metz et d’Evry. Répondez en francais svp (Compare the production of the Metz and d’Every sites. Respond in french, please)
This prompt — as well as a data scheme and possible small data sample — is sent to an LLM via API. The model’s provided response is derived from the entirety of available data, ensuring constant relevancy and an unlimited scope, regardless of the quantity of existing data.
Considering the limited amount of data sent out and the possibility to anonymize any sensitive data, the public version of the API could be used. Tighter control on data and input could be achieved with a containerized version of the LLM.
For the responsible use of an LLM to accelerate self-service analytics on CO2 forecasting, the organization should have an overarching Responsible AI policy to enforce consistent practices. In addition:
- As the LLM is used to generate insight via a chatbot approach, it is important that the insights provided by the LLM are marked as AI generated and that end users know they are interacting with an AI system.
- Limitations of the model should be documented, and end users should be encouraged to use best judgment when working with outputs of the model. In particular when dealing with issues that could impact human safety concerns, teams should pay attention to the error rate of the underlying model.
- To address transparency and potential limitations of the LLM/chatbot’s responses, a panel titled “Data Sources Used for the Insights” provides users with an understanding of which columns from which datasets were used to generate the insights.
- As answers are delivered on the fly, retroactive auditability is not enabled, and exact answers could vary based on data and behavior of the LLM. Training of the teams to advise on scope of usage is strongly advised.