Choose one track. Master’s proposals target Spring 2027. Final scope and programme approval are agreed with the supervisor; bachelor scopes are suggested adaptations.
TRACK 01 · IMPLEMENTATION INCLUDED
Technical track
Develop a working solution and test whether it addresses the problem.
Possible research question
Which verification and presentation mechanisms help developers find incorrect AI-generated changes without disproportionate review work?
Suggested tasks
Read research on review of AI-generated software and compare existing solutions.
Identify one problem faced by software reviewers.
Write a research question and define what the solution should do.
Choose a few automated checks for code quality, tests and security.
Build a review screen that shows which checks passed, which failed and what has not been checked.
Compare it with a normal AI-assisted code review. Measure defects found, review time and mistakes that are missed.
Explain what worked, what did not, and how the results compare with earlier research.
Evaluation, degree scope and deliverables
Study and evaluation
Compare the implemented solution with a normal AI-assisted code-review workflow. Combine reproducible technical tests with an appropriate empirical evaluation.
Defects accepted and useful defects found
False alarms
Total review-plus-rework time
Degree scope
Bachelor’s
Build a review dashboard for one language and a small defect set. Connect existing tools and evaluate whether users can identify seeded errors.
Master’s
Compare evidence presentation across controlled changes, measuring completed review and rework rather than first-draft generation speed.
Background
Programming
Automated tests and Git
Basic secure-development practices
Possible deliverables
A focused literature review, justified problem and research question
A working prototype with source code and setup instructions
A reproducible comparison and an appropriate study of use
A report explaining design lessons, results and limitations
TRACK 02 · NO IMPLEMENTATION REQUIRED
Non-technical track
Study existing systems, information or work practices. You do not need to develop software.
Possible research question
How do developers decide whether AI-generated changes are ready to accept, and what evidence do they rely on?
Suggested tasks
Read earlier studies of review of AI-generated software.
Choose one problem and write a research question the study can answer.
Observe reviews of prepared changes or authorised existing pull requests.
Interview developers about trust, review responsibility and the meaning of passing automated checks.
Compare review reasoning across evidence packages containing known defects and explicit verification limits.
Analyse the interviews, observations or documents using a clearly described method. Look for disagreements as well as common patterns.
Explain the findings, compare them with earlier research and suggest practical improvements.
Evaluation, degree scope and deliverables
Study and evaluation
Use a bounded empirical study of review of AI-generated software. Justify case selection, recruitment and the analysis method. Distinguish observed behaviour from participants’ perceptions; use triangulation or a comparison where it serves the research question.
Review strategies and defects overlooked in bounded cases
Trust, accountability and review-plus-rework effort
Evidence for the findings, conflicting cases and limits of the study
Degree scope
Bachelor’s
Study one case or a small set of existing materials. Agree the interviews, documents or scenario tasks with the supervisor. Describe the method, analyse the findings and give practical recommendations.
Master’s
Use a clear research question and relevant IS theory. Justify the cases, participants and analysis method. Explain what the findings add to earlier research and where they may apply. No software development is required.
Background
Literature review and academic writing
Qualitative or quantitative research methods
Interest in review of AI-generated software; no programming prerequisite
Possible deliverables
A literature review and research question
A study plan and approved research material
An analysis supported by interviews, observations, documents or scenario results
A thesis with findings, recommendations and limitations
Scope and access
Keep evaluation defects and tests independent of generation where possible. Passing automated checks must not be presented as proof of correctness. These implementation-related limits apply when developing or testing a technical solution. For a non-technical study, agree access to participants or existing materials early, use approved or fictional cases where appropriate, and distinguish perceptions from observed outcomes.
Agree access to data, participants or existing materials and any required ethics or privacy review before committing. A non-technical track needs a systematic study, not a working prototype.
Full academic proposal
Working topic
Evidence-Based Human Review of AI-Generated Code Changes
Brief outline
This proposal examines review of AI-generated software in the work and information needs of software reviewers. The technical track combines a literature review and justified gap with requirements, design, implementation and evaluation of a bounded solution. The non-technical track investigates practices, experiences or organisational conditions through a systematic study of existing systems, documents or scenarios, without requiring implementation. Choose one track and agree the final research question, degree scope and contribution with the supervisor.
Programme fit
Information Systems. These are suggested research approaches, not a statement of confirmed programme policy. Agree the final title, track, degree scope and contribution with the supervisor and programme.
Shared research foundation
Review the literature; identify and justify a gap; formulate research questions; conduct a systematic study; analyse the evidence; explain the contribution relative to prior research and discuss limitations. The technical track additionally includes requirements, design, implementation and evaluation of an artifact.
