At Enerdata, Artificial Intelligence is deployed as a strategically integrated lever to reinforce our expertise and the quality of our data. Our approach is strictly focused on control, security, and efficiency, ensuring that AI complements, without ever replacing, the know-how of our experts.
AI serves as an additional tool within our process chain, facilitating the selection, processing, analysis, and translation of massive data streams. This allows our domain experts to dedicate their efforts on interpretation and strategic value creation.
Our commitment to data confidentiality and frugal AI is actualised through the development and on-site hosting of our own LLM (Large Language Model) tool, named "OSE". Based on the high-performing Mistral AI models, this choice enables us to guarantee the confidentiality of our sensitive data and to better control usage. This is essential for meeting our commitments regarding energy consumption and digital sufficiency.
The internal deployment of OSE also allows us to specialise and use these models for workflows specific to our business, without any data externalisation, whether it involves assisting with repetitive or high-value tasks. Furthermore, collaborating with a national partner like Mistral AI ensures our compliance with European regulations, notably the AI Act and the GDPR, while also benefiting from greater transparency regarding actual energy costs. Access to open-source models is also vital, enabling us to train and specialise them for precise data applications, such as the development of an internal chatbot for navigating our database platform.
In addition to Generative AI, we rely on Machine Learning and Deep Learning models to enhance and guarantee the comprehensiveness and relevance of our energy data, particularly for refining our forecasts through Nowcasting. Nowcasting is used to perform advanced forecasting of missing data from the previous and current year for the energy data of the areas studied.
Our models specifically enable us to estimate daily consumption of gas and electricity to feed our power price projections model. These projections are based on data such as national calendars, meteorological data, and long-term trends provided by our EnerFuture scenarios.
Recently, we have implemented our own AI agent chains, combining open-source classification models and our internal OSE LLM to optimise several critical operational processes:
- Monitoring and processing sectoral intelligence.
- Daily selection and classification of tenders.
- Supporting in analysis drafting, particularly by producing news summaries on Energy actors, sourced from Internet searches.
Extracting complex data, whether sourced from APIs, unstructured documents, or charts.