Thesis work is an excellent way to get closer to Scania and build relationships for the future. Many of today's employees began their Scania career with their degree project. Choose one of four thesis topics exploring how safety can be analysed, assured, and maintained in commercial vehicle operations. Topics range from AI-assisted incident analysis and modular safety cases to fallback strategies and motion planning under reduced braking capability. Please indicate your preference regarding your chosen topic in the pre-screening question. MSc Thesis Proposal 1 Agent-Assisted Safety Pattern Detection in Test Vehicle Operations Background and Motivation Semiannual reviews of test-vehicle incidents must combine records from different projects and reporting periods. Inconsistent categories, duplicates, and missing information make recurring safety problems difficult to identify. Repeating the analysis manually also consumes engineering time. An AI agent could execute a defined analysis workflow repeatedly, but its results must remain reproducible, traceable, and subject to engineer approval. Objective Develop and evaluate a reusable AI agent skill that prepares incident records, runs a defined pattern-detection workflow, and produces an evidence-linked draft report for each reporting period. Can an agent-executed workflow reduce preparation effort while preserving the correctness and traceability of a deterministic analysis pipeline? Which incident patterns add useful information beyond category counts, and how robust are they to missing, duplicated, or inconsistent records? Job description Review incident-pattern methods and define a common CSV/Excel record structure using synthetic records representing two or three reporting periods. Implement a deterministic Python pipeline for schema validation, duplicate detection, missing-value flags, category counts, and one interpretable pattern-detection method. Define an executable AI agent skill with explicit inputs, tools, ordered steps, output schema, and stop conditions. The agent invokes the pipeline, flags ambiguous records, and drafts a report with links to supporting records. Add human approval for data corrections and safety conclusions; prohibit unsupported imputation, invented incidents, and autonomous release decisions. Log agent actions and preserve reproducible outputs. Demonstrate reuse on a previously unseen reporting period without modifying the agent skill. Compare agent execution with a manual workflow and the standalone deterministic pipeline; measure record-level accuracy, unsupported statements, traceability, run-to-run consistency, analyst time, required human intervention, and pattern stability under controlled data-quality changes. Expected Outcomes A versioned agent skill specification, reusable Python analysis code, synthetic test datasets, evidence-linked period reports, and a quantitative comparison. The contribution is an evaluated method for repeatable, human-supervised agent
