SmartSOP
AI-supported frameworks for intelligent SOP generation
Start: 03/2026
End: 02/2028
SmartSOP develops an AI-supported framework for the automated generation, structuring, validation and traceable documentation of standard operating procedures. The project aims to transform unstructured scientific and technical sources, standards, method descriptions and application documents into machine-readable and auditable SOP structures. It addresses a key digitalization gap in regulated laboratory and quality processes: today, SOPs are often stored as static PDF, Word or LIMS documents, offer limited interoperability and require substantial manual effort to maintain. SmartSOP combines domain-specific large language models, retrieval-augmented generation, semantic segmentation, ontologies, validation rules and traceability mechanisms into an integrated toolchain. The generated SOPs are intended to be understandable for humans while also being processable by machines. This enables laboratory processes to become more transparent, reproducible and easier to integrate into digital quality management, laboratory and instrument systems.
In SmartSOP, Angi GmbH, IUTA e. V. and FZI jointly develop a modular software framework for the intelligent generation and use of SOPs. The project starts from the observation that analytical methods are widely available in scientific publications, standards, manufacturer documentation, and internal laboratory documents, but rarely in a form that is directly reusable, machine-readable, or auditable. SmartSOP introduces a new technical approach: relevant text sources are collected, preprocessed, semantically segmented, and transformed into structured SOP modules using domain-adapted AI models. These modules are then validated against a generic SOP data model, enriched with metadata, source references and confidence scores, and provided in standardized formats such as JSON or XML. The initial focus is on liquid chromatography coupled with mass spectrometry, while the modular architecture is intended to support transfer to other analytical methods. The project also develops mechanisms for version control, change tracking, audit-trail generation, and workflow automation. The result is a prototypical end-to-end system that connects literature and method sources with validatable, machine-compatible SOPs.
FZI is responsible for the AI-based interpretation, generation, validation, and maintenance of machine-readable SOPs. FZI's main task is to develop an AI backend that identifies SOP-relevant information in heterogeneous technical sources, structures it semantically, and maps it to the shared SOP data model. FZI adapts open-weight language models to the laboratory and SOP domain using in-context learning, prompting strategies and, where appropriate, LoRA or QLoRA fine-tuning. Retrieval-augmented generation is used to ground outputs in versioned and verified knowledge sources and to reduce hallucinations. Another focus is the development of confidence-scoring methods, for example, based on logits, probes, self-consistency, or cross-model checks. FZI also develops mechanisms for semantic comparison of SOP versions and for the automated extension of audit-trail entries. In later work packages, FZI integrates the AI components into a robust API layer, supports the generation of standardized laboratory workflows, and optimizes semantic workflows based on internal laboratory and integration tests.