Swiss AI Center Presentations
Week-Ahead Drinking Water Demand Prediction Using Machine Learning: A Case Study from Val de Bagnes
Fanny Terrettaz, Emilie Neveu (HES-SO Valais-Wallis)
Developed within the STREAM partnership between HES-SO Valais and BlueArk, this research addresses operational needs expressed by Altis, a water utility operator seeking improved demand forecasting capabilities for supply optimization and infrastructure planning in an Alpine municipality. Using smart meter data from Val de Bagnes (2022-2024), we developed machine learning models to predict hourly drinking water demand one week ahead. Multiple algorithms were tested; LightGBM, using calendar and temporal features (such as day of the week and last week's consumption values) along with meteorological variables (temperature, precipitation, daylight duration), emerged as the best performer. LightGBM was trained on data from 2022-01-01 to 2024-10-31 and tested on the last two months of 2024. It predicted the average consumption per smart meter with a MSE of 1.59×10⁻³ m³/h (November 2024) and 2.03×10⁻³ m³/h (December 2024), outperforming the baseline persistence model (last week's values: 2.1×10⁻³ m³/h and 2.98×10⁻³ m³/h respectively). The model captures seasonal variations effectively, including robust performance during December's holiday period. Interestingly, meteorological variables emerged as non-significant predictors compared to historical consumption patterns (lagged consumption values at hourly, daily, weekly, and annual timescales). However, consumption peaks are underestimated, and further research is needed to improve this. This work demonstrates practical ML applications for municipal water management. Future developments include extending the approach to other water sectors, integrating irrigation demand predictions for coordinated supply management, and deploying the system to improve operational decision-making at Altis facilities.
AI-Driven Summaries of Parliamentary Sessions: Enabling SME-Ready Policy Intelligence in Valais
Sébastien Gard (HES-SO Valais-Wallis)
Small and medium-sized enterprises (SMEs) often lack the resources to monitor complex legislative activity, yet political decisions strongly impact their operations. The canton of Valais produces hundreds of pages of parliamentary session documents per sitting, making timely access to actionable information a challenge. In collaboration with FER Valais, HES-SO Valais applied an Innovation Cheque-backed initiative to develop an AI-based system that automates data extraction, reasoning and targeted summarisation of parliamentary sessions, delivering insights tailored to SMEs' needs. The solution ingests raw PDF documents from the Valais Parliament, aligns them with planning data, cleans and structures the content, then applies a custom prompt-driven reasoning model via the OpenAI API. The system identifies key themes (economic, regulatory, fiscal), assesses potential SME impact and produces concise, structured summaries geared to business and advocacy users. All design choices reflect industry-academia collaboration and practical deployment in a real-world public-sector context. This case study aligns with the conference's emphasis on AI implementation in SMEs and industry-academia partnerships. Its practical relevance is clear: the tool transforms hours of manual reading into near-real-time insights, offering a reproducible model for other regions, public-policy watchers or business federations. On innovation, the approach combines domain-specific prompt engineering, document-workflow automation and stakeholder-centric output design. Results include a dramatic reduction in processing time, consistent summary quality and a scalable architecture ready for extension and replication. Looking ahead, the project will extend to automated vote-analysis, trend-dashboards and multi-canton adaptation—emphasising lessons learned and a roadmap for service-oriented AI deployment. This makes the work broadly applicable, replicable and directly aligned with the evaluation criteria: relevance to other organisations, originality of approach and measurable outcomes.