An AI model is a trained mathematical system that processes inputs and produces predictions, classifications or generated content. Well-known examples are GPT-4, Claude, Gemini or Llama for language, Stable Diffusion or Imagen for images, Whisper for speech transcription.
Building a model involves several steps: data collection and curation, architecture choice (such as transformers), training on large datasets, fine-tuning for specific tasks, safety checks and deployment. Effort ranges from small research projects to investments in the hundreds of millions.
For users, the distinction between open-weight models (often with open licences) and proprietary models (API access, terms of use, data and security rules) matters. In Germany, GDPR and the EU AI Act are particularly relevant for the choice.
Lunigi combines several models: commercial LLMs for research and summarisation plus dedicated domain-specific models for matching and AI-safety evaluation.