Small apps built with AI — Technical https://thesis.uya.no/proposals/build-it-with-guardrails/#technical-track BACHELOR Final result: A working app, access tests and handover documentation. Task: Build one administrative app from the fixed specification using an existing authentication helper. Data: 12 approval-app specs × 2 versions (planned) Choose one leave, equipment or expense specification from the proposed set. Define employee/manager/admin permissions and 10 checks: 5 workflow and 5 authorization checks. Requires: administrative users, a requirements reviewer, isolated execution and inference compute. Method: Verify workflow and permission requirements with scripted tests and administrator walkthroughs. Plan 4 participant sessions. STEPS Build and test 1. Read the starting sources and choose one established implementation method. 2. Write the requirements, data fields, system diagram and test cases. 3. Prepare 12 approval-app specs × 2 versions. Make the answer key and pass criteria before testing. 4. Build a working version of a working app, access tests and handover documentation. 5. Run function, integration and failure-case tests. Record each result. 6. Run the practical evaluation and list the changes the system still needs. MASTER Final result: App generator, 12 specs and a 24-app comparison. Research question: Investigate whether a fixed authorization scaffold changes the kinds of defects in AI-built apps. Hold model/specification/budget constant; compare scaffolded and generated authorization, then connect test failures to administrator handover decisions. Research result: Measured security, correctness and ownership results for constrained AI app development. STEPS Design, build and test (DSR) 1. Read the newest papers and list the closest existing systems. 2. Write down the versions, fields, data split, case assignment, random seeds and correct answers for 12 approval-app specs × 2 versions. 3. Draw the user workflow, data model and system architecture. List the requirements and pass criteria. 4. Build a working version of app generator, 12 specs and a 24-app comparison. 5. Test every function, connection and failure case. Save the failed tests as well as the passed tests. 6. Compare the system with the named alternative. Then run the user task or decision task in the assignment. 7. Report the measured result, the failed cases and the design lessons another team can reuse. CURRENT PROJECT LITERATURE A systematic literature review on the impact of AI models on the security of code generation (2024, systematic review): https://doi.org/10.3389/fdata.2024.1386720 Synthesizes current security risks and review needs for AI-generated code. Balancing Security and Correctness in Code Generation: An Empirical Study on Commercial Large Language Models (2025, peer-reviewed): https://doi.org/10.1109/TETCI.2024.3446695 Separates functional correctness from vulnerability outcomes in generated code. IS THEORY STARTING POINTS Liu et al. (2023) — Is Your Code Generated by ChatGPT Really Correct?: https://proceedings.neurips.cc/paper_files/paper/2023/file/43e9d647ccd3e4b7b5baab53f0368686-Paper-Conference.pdf Distinguish passing a public test suite from independently assessed correctness. Nissenbaum (2004) — Privacy as Contextual Integrity: https://digitalcommons.law.uw.edu/wlr/vol79/iss1/10/ Analyse recipient, purpose and information flow rather than treating privacy as a role label. Search Scopus or Web of Science and ACM Digital Library using the topic query, then follow citations to the thesis start date. Record searches and compare methods, data, findings and limitations in literature-matrix.csv. Use that review to confirm or revise the gap and choose a current comparator. The linked papers are starting points. PILOT QUESTION Do explicit schemas and access rules improve the handover quality of AI-generated approval apps? EXAMPLE TOOLS Qwen2.5-Coder-7B-Instruct, Flask, SQLite, pytest Equivalent tools are fine. FIRST THREE ACTIONS 1. Write one equipment-request specification and its 10 acceptance checks before generating an app. Confirm access to inference compute and record model, software and hardware versions. 2. finalise the same Flask/SQLite template, model and decoding settings for both conditions. 3. Generate both versions in isolation and validate the permission tests before expanding to 12 specifications. LITERATURE SEARCH end user development AI generated applications governance maintenance DATA 12 approval-app specs × 2 versions (planned) 4 leave, 4 equipment, 4 expense workflows; employee/manager/admin roles. Define 10 checks per spec before generation: 5 workflow and 5 authorization checks. Qwen2.5-Coder-7B-Instruct: https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct Transformers generation settings: https://huggingface.co/docs/transformers/main_classes/text_generation HOW TO TEST IT Use fixed Qwen Coder revisions and greedy settings; run direct-request checks in isolation. Count workflow passes, prohibited actions and failed builds separately. Four administrative users inspect sampled workflows and explain handover requirements. FINAL RESULT App generator, 12 specs and a 24-app comparison. ACCESS OR PEOPLE 4 administrative users, a requirements reviewer, isolated execution and inference compute. TEMPLATES https://thesis.uya.no/starters/build-it-with-guardrails/technical/data-plan.csv https://thesis.uya.no/starters/build-it-with-guardrails/technical/evaluation.csv https://thesis.uya.no/starters/build-it-with-guardrails/technical/literature-matrix.csv Use participant codes instead of names or email addresses.