Master Thesis in Adaptive Architecture for Human-AI Annotation Systems

This thesis project is for two Master's students working together. You will design, implement and evaluate a confidence-aware verification architecture for an existing annotation tool. The focus is on software architecture, integration, testing, maintainability, traceability, fallback behaviour and effective human oversight. Developing new AI perception models is not the primary objective.

The work may include designing modular interfaces for confidence providers, verification policies, decision logging and annotator feedback; implementing and comparing threshold-based and learned routing policies; integrating attention cues; and recording routing decisions and human outcomes. You will also develop automated tests, benchmark the adaptive architecture against existing annotation workflows, and evaluate maintainability through change scenarios. A counterbalanced user study may examine verification behaviour, cognitive workload, calibration and automation bias. Depending on feasibility, additional confidence signals such as out-of-distribution detection or scene complexity may be explored.

You will join the Verification & Validation (V&V) Virtual Driver team at V.A.S., which works with software integrity for autonomous decision-making and control. The team combines in-office development with in-vehicle testing.

You are a final-year Master's student in Software Engineering, Computer Science or a related field. You have a strong interest in software architecture, software quality and maintainable systems, and are comfortable programming in Python. A foundation in modular software design, interface design, automated testing and version control is expected. Experience or interest in AI-assisted systems, machine learning, computer vision or human-computer interaction is advantageous, but prior specialisation in machine learning or object detection is not required.

The thesis provides practical experience in architectural design, component integration, automated testing and empirical evaluation of a safety-critical human-AI system. Preferred start date is January 2027. The application deadline is 15 November 2026.