LLM-based JSON Mapping and Blockchain Integration for Digital Product Passports

The European Commission envisages Digital Product Passports (DPPs) as a mechanism to enable traceable, transparent, and standardized product data across supply chains. This work presents a modular pipeline that transforms raw sensor data into verifiable DPP records using large language models (LLMs) for data standardization and blockchain technology for tamper-proof storage. The system maps unstructured machine-level data to a standardized JSON format and stores it immutably on the Waves blockchain via smart contracts, thereby enabling auditable, machine-readable records suitable for regulatory use. A novel evaluation dataset is introduced to simulate daily production scenarios with varying mapping complexity. The performance of the system is assessed using both proprietary and open-weight LLMs. Results show that the proprietary model achieves the highest accuracy and lowest latency, while open-weight models perform worse as input complexity increases. Multiple prompting strategies were compared, revealing that direct mapping, via few-shot or zero-shot prompts, consistently delivered higher accuracy than approaches based on generating transformation functions. Structured output formatting was also assessed: While it ensured schema validity, it often compromised mapping reliability by introducing incorrect values, likely due to disruptions in model reasoning from output constraints. The proposed architecture demonstrates reliable end-to-end operation with low latency and is suitable for batch-level deployment in real-world production environments. From a practical perspective, the results clarify trade-offs between model choice, prompting strategy, and operational reliability in automated DPP generation. For policymakers, the findings highlight how choices around schema clarity and data granularity shape system design and operational effort in future DPP implementations.