{"id":"W7125580550","doi":"10.1109/cascon66301.2025.00069","title":"Can AI Build Systems? an Exploratory Study on Generating Software Architecture With LLMS","year":2025,"lang":"","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Blueprint; Software architecture; Architecture; Software architecture description; Reference architecture; Software development; Resource-oriented architecture","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006151986,0.0005196349,0.0002591911,0.0007972281,0.000726752,0.001659362,0.001150288,0.0009571193,0.00267752],"category_scores_gemma":[0.0375243,0.0004274525,0.0004459468,0.0006436887,0.0017589,0.004693819,0.001528743,0.001537028,0.0004000396],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001337507,"about_ca_system_score_gemma":0.000968243,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003002741,"about_ca_topic_score_gemma":0.005347399,"domain_scores_codex":[0.9964297,0.002717343,0.00009250036,0.0002613088,0.0004045929,0.00009453414],"domain_scores_gemma":[0.9575946,0.03695884,0.0008925255,0.0030362,0.001182631,0.0003351331],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001720351,0.003437679,0.07584152,0.002163456,0.0001429191,0.002339479,0.1094408,0.2037834,0.03938284,0.1310849,0.007641283,0.4230215],"study_design_scores_gemma":[0.0004243648,0.00401436,0.03268161,0.0004402233,0.0001279458,0.001033345,0.04104831,0.7169549,0.03422904,0.09872664,0.07014732,0.0001720164],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9103428,0.0001906375,0.07447782,0.00108918,0.0000198845,0.0005178201,0.0001918419,0.0009454837,0.01222464],"genre_scores_gemma":[0.9145621,0.0001470431,0.08275456,0.0002055521,0.000007691266,0.0002236035,0.0002793229,0.000151499,0.001668462],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006151986,"threshold_uncertainty_score":0.0325352,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01849062765235339,"score_gpt":0.2764884218478362,"score_spread":0.2579977941954829,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}