Project

Full title: Benefits on using LLMs on STRADA

Purpose and goal

STRADA is a world-leading road accident database central to Sweden’s "Vision Zero." However, the reporting process can be cumbersome for contributors. This pre-study investigates how Large Language Models (LLMs) can optimize the system. We will assess the risks and rewards of using AI to improve historical data and simplify new entries, ultimately increasing the database's value for both researchers and practitioners.

Planned approach and activities

The pre-study builds on the momentum of previous SAFER workshops, which identified five key themes for AI integration in road safety data. While long-term implementation requires a multi-year effort, this pre-study serves as a strategic catalyst to define the requirements and technical and organizational feasibility. This pre-study will prioritize the Technical and Workflow Foundation  through the following activities:

• State-of-the-Art Review: A targeted literature study on LLM applications for structuring narrative accident reports and "sketches-to-data" conversion, specifically focusing on Swedish-language models. Evaluating 3-5 models based on efficiency, license, and cost.
• Feasibility Pilot (Themes A & B): A small-scale technical "sandbox" test to evaluate LLM accuracy in extracting structured variables from existing free-text entries (from the sample of ~50 accidents). The aim is to validate the feasibility to justify a larger project.
• Stakeholder Requirements Mapping (Theme E): Collaborative sessions with Transportstyrelsen and data providers to identify the highest-impact "friction points" in current workflows. Provide an overview/map of data providers and bottlenecks in the process.
• Risk & Dependency Assessment: Create a matrix of challenges (legal, validation, trust, and data structure) identified by SAFER partners where future ethically and technically sound AI tools can be used.

Expected results

The pre-study will deliver a Technical and Financial proposal for a larger implementation effort. Results will be synthesized into a formal project proposal, highlighting clear opportunities for innovation, technical risks, and data-security challenges. Additionally, the project will produce a curated list of relevant funding calls to ensure the long-term sustainability of the initiative.