AI meets ROI
Lunar is a European tech lab focused on the next generation of foundation models. Combining extensive strategic and technology experience, we guide our clients to translate the latest foundation models into long-term value for their organisations, where we provide both the strategic guidance as well as the finetuning of the models to deliver the actual value.
About foundation models
Foundation models (FMs) – also known as base models - are advanced, multi-purpose AI models trained on vast and diverse datasets. Where FMs trained on language data are well-known to the public (e.g., ChatGPT), FMs trained on other data types are still in their early days. Given their typically diverse input data, FMs provide unique advantages such as i) more cross-domain predictions, ii) faster time-to-market, and iii) potentially higher accuracy than traditional models – particularly for predictions made in low-data environments. Due to the rapidly declining token cost for FMs as well as the explosion of the amount of FMs available, many use cases beyond language are becoming economically viable.
Meet Geo-FMs (GFMs)
Geospatial foundation models (GFMs) are typically trained on satellite imagery, climate data, land use data, demographic, topographic and sensor data. Geo-FMs are therefore invaluable for all predictions involving localisation such as land-use predictions (e.g. ideal crop transitions, optimal locations for renewables), land degradation (e.g., deforestation, desertification), supply chain risk assessments, environmental risk assessments, and smart city design. Learn more.
Meet Time-FMs (TFMs)
Time-series Foundation Models (TFMs) are typically trained on time-dependent data, such as economic data (e.g. stock prices, exchange rates, rate of inflation, GDP growth), weather data, utilities data (e.g. electricity demand), healthcare (e.g. disease outbreaks, hospital admissions), and commercial data (e.g. retail, e-commerce). Use cases for these models can therefore vary from demand forecasting to climate modelling (e.g. carbon or pollution levels). Learn more.
Meet Urban FMs (UFMs)
Urban Foundation Models (UFMs) are a new category of models revolutionizing how we do our urban planning, typically trained on climate and weather data, traffic flow data, drone data, and geospatial data. Use cases include land-use prediction (residential, commercial), urban resilience (disaster support), route optimisation, public transport optimisation, traffic flow analysis, crime hotspot prediction, renewable energy predictions (solar installations), air quality and urban heat island predictions. Learn more.
Meet Material FMs (MFMs)
Material Foundation Models (MFMs) are typically trained on material property databases. MFMs can be used for a variety of predictions in both biological (protein structures, DNA structure, RNA structures, gene expression etc.) and synthetic materials (sustainable/new materials discovery, performance optimisation of materials, predicting material failure). Learn more.
