{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T20:09:12Z","timestamp":1792526952882,"version":"4.5.0"},"reference-count":51,"publisher":"Wiley","issue":"8","license":[{"start":{"date-parts":[[2026,7,6]],"date-time":"2026-07-06T00:00:00Z","timestamp":1783296000000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/onlinelibrary.wiley.com\/termsAndConditions#vor"},{"start":{"date-parts":[[2026,7,6]],"date-time":"2026-07-06T00:00:00Z","timestamp":1783296000000},"content-version":"tdm","delay-in-days":0,"URL":"http:\/\/doi.wiley.com\/10.1002\/tdm_license_1.1"}],"funder":[{"DOI":"10.13039\/100031060","name":"European High Performance Computing Joint Undertaking","doi-asserted-by":"publisher","award":["EHPC\u2010AI\u20102024A06\u2010008"],"award-info":[{"award-number":["EHPC\u2010AI\u20102024A06\u2010008"]}],"id":[{"id":"10.13039\/100031060","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["Expert Systems"],"published-print":{"date-parts":[[2026,8]]},"abstract":"<jats:title>ABSTRACT<\/jats:title>\n                  <jats:p>Automating multi\u2010stage decision workflows in regulated industrial domains demands intelligent systems capable of orchestrating heterogeneous evidence sources, such as uncontrolled photographs, free\u2010text narratives and structured forms, under strict operational and compliance constraints. We present FADES, a production\u2010grade expert system that integrates conversational AI with computer vision to automate end\u2010to\u2010end claim evidence interpretation and vehicle damage assessment. The system addresses three core challenges: structured data acquisition via state\u2010machine\u2010controlled conversational agents powered by tool\u2010enabled Large Language Models (LLMs); robust damage localization through modern object detection; and deterministic severity estimation via panel\u2010aware geometric reasoning and rule\u2010based logic. Our methodology combines backend\u2010enforced slot\u2010filling for First Notice of Loss (FNOL) data collection, ensuring compliance with the Italian CAI standard, with a multi\u2010stage damage analysis pipeline that reconciles detection outputs with vehicle panel segmentation. We systematically evaluated diffusion\u2010based (DiffusionDet) and transformer\u2010based (RF\u2010DETR, D\u2010FINE) detectors on 13,353 real accident images comprising 41,057 annotations across four damage classes. RF\u2010DETR Large achieves  ( over DiffusionDet),  () and , while satisfying real\u2010time processing requirements. Operational validation demonstrates that the integrated system reduces claim processing time from 30 to 12\u2009min, achieving time\u2010to\u2010completion under 7\u2009min at a sustained throughput of 45\u2009cycles per day. Cross\u2010dataset experiments reveal substantial domain gaps ( on CarDD), confirming the necessity of insurance\u2010specific training data. Micro\u2010ablation analysis shows full pipeline integration outperforms partial implementations by  in end\u2010to\u2010end efficiency. These results provide empirical evidence that hybrid deterministic\u2010generative architectures deliver the reliability and auditability required for regulated industrial applications.<\/jats:p>","DOI":"10.1111\/exsy.70340","type":"journal-article","created":{"date-parts":[[2026,7,6]],"date-time":"2026-07-06T08:40:10Z","timestamp":1783327210000},"update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":2,"title":["<scp>Integrating LLMs With Computer Vision: A Geometrically Invariant System for First Notice of Loss in Car Insurance<\/scp>"],"prefix":"10.1111","volume":"43","author":[{"given":"Vito","family":"Arconzo","sequence":"first","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Gerardo","family":"Gorga","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Gonzalo","family":"Gutierrez","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Gabriel","family":"Pinos","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Simone","family":"Poggi","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Meher Anvesh","family":"Rangisetty","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-3809-8886","authenticated-orcid":false,"given":"Lorenzo","family":"Ricciardi Celsi","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Federico","family":"Santini","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Enrico","family":"Scianaro","sequence":"additional","affiliation":[{"name":"Insoore \u2013 Whoosnap Srl  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-1963-3548","authenticated-orcid":false,"given":"Indro","family":"Spinelli","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Perception and Intelligence Laboratory (PINlab) Sapienza Universit\u00e0 di Roma  Roma Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"311","published-online":{"date-parts":[[2026,7,6]]},"reference":[{"key":"e_1_2_9_2_1","unstructured":"Achiam J. S.Adler S.Agarwal et\u00a0al.2023.\u201cGPT\u20104 Technical Report.\u201d arXiv Preprint arXiv:2303.08774.https:\/\/arxiv.org\/abs\/2303.08774."},{"key":"e_1_2_9_3_1","first-page":"19","volume-title":"Enabling the Digitalization of Claim Management in the Insurance Value Chain Through AI\u2010Based Prototypes: The ELIS Innovation Hub Approach","author":"Andreozzi A.","year":"2021"},{"key":"e_1_2_9_4_1","doi-asserted-by":"publisher","DOI":"10.3390\/electronics14071362"},{"key":"e_1_2_9_5_1","doi-asserted-by":"publisher","DOI":"10.3390\/electronics13224333"},{"key":"e_1_2_9_6_1","unstructured":"Bdeo.2024.\u201cVisual Intelligence for Motor and Home Insurance Companies.\u201dhttps:\/\/bdeo.io\/en\/."},{"key":"e_1_2_9_7_1","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00976"},{"key":"e_1_2_9_8_1","first-page":"1877","volume-title":"Advances in Neural Information Processing Systems","author":"Brown T.","year":"2020"},{"key":"e_1_2_9_9_1","unstructured":"CAI M.2026.\u201cConstatazione Amichevole Di Incidente.\u201dhttps:\/\/www.carrozzeriaparadiso.it\/modulo\u2010cai.pdf."},{"key":"e_1_2_9_10_1","doi-asserted-by":"crossref","unstructured":"Carion N. F.Massa G.Synnaeve N.Usunier A.Kirillov andS.Zagoruyko.2020.\u201cEnd\u2010to\u2010End Object Detection With Transformers.\u201dhttps:\/\/arxiv.org\/abs\/2005.12872.","DOI":"10.1007\/978-3-030-58452-8_13"},{"key":"e_1_2_9_11_1","doi-asserted-by":"crossref","unstructured":"Chen S. P.Sun Y.Song andP.Luo.2023a.\u201cDiffusiondet: Diffusion Model for Object Detection.\u201dhttps:\/\/arxiv.org\/abs\/2211.09788.","DOI":"10.1109\/ICCV51070.2023.01816"},{"key":"e_1_2_9_12_1","doi-asserted-by":"crossref","unstructured":"Chen S. P.Sun Y.Song andP.Luo.2023b.\u201cDiffusiondet: Diffusion Model for Object Detection\u2014Backbones.\u201dhttps:\/\/github.com\/ShoufaChen\/DiffusionDet\/blob\/main\/configs\/diffdet.coco.swinbase.yaml.","DOI":"10.1109\/ICCV51070.2023.01816"},{"key":"e_1_2_9_13_1","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00352"},{"key":"e_1_2_9_14_1","doi-asserted-by":"crossref","unstructured":"Dhieb N. H.Ghazzai H.Besbes andY.Massoud.2019.\u201cA Very Deep Transfer Learning Model for Vehicle Damage Detection and Localization.\u201d 158\u2013161.","DOI":"10.1109\/ICM48031.2019.9021687"},{"key":"e_1_2_9_15_1","unstructured":"Dubey A. A.Jauhri A.Pandey et\u00a0al.2024.\u201cThe Llama 3 Herd of Models.\u201d arXiv Preprint arXiv:2407.21783.https:\/\/arxiv.org\/abs\/2407.21783."},{"key":"e_1_2_9_16_1","unstructured":"European Parliament and Council.2016.\u201cRegulation (EU) 2016\/679 of the European Parliament and of the Council of 27 April 2016.\u201dhttps:\/\/gdpr\u2010info.eu\/."},{"key":"e_1_2_9_17_1","unstructured":"Gal Y. andZ.Ghahramani.2016.\u201cDropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning.\u201dhttps:\/\/arxiv.org\/abs\/1506.02142."},{"key":"e_1_2_9_18_1","first-page":"1180","volume-title":"International Conference on Machine Learning","author":"Ganin Y.","year":"2015"},{"key":"e_1_2_9_19_1","unstructured":"Google Cloud.2024.\u201cVertex AI Documentation: Data Residency and Privacy.\u201dhttps:\/\/cloud.google.com\/vertex\u2010ai\/docs\/generative\u2010ai\/data\u2010privacy."},{"key":"e_1_2_9_20_1","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2024.3506563"},{"key":"e_1_2_9_21_1","unstructured":"Jacqmin L. L. M.Rojas\u2010Barahona andB.Favre.2022.\u201cA Recent Survey on Dialogue State Tracking: Models Datasets and Challenges.\u201d arXiv Preprint arXiv:2003.01390.https:\/\/arxiv.org\/abs\/2003.01390."},{"key":"e_1_2_9_22_1","doi-asserted-by":"publisher","DOI":"10.1145\/3571730"},{"key":"e_1_2_9_23_1","unstructured":"Jiang A. Q. A.Sablayrolles A.Mensch et\u00a0al.2023.\u201cMistral 7b.\u201d arXiv Preprint arXiv:2310.06825.https:\/\/arxiv.org\/abs\/2310.06825."},{"key":"e_1_2_9_24_1","unstructured":"Khanam R. andM.Hussain.2024.\u201cYolov11: An Overview of the Key Architectural Enhancements.\u201dhttps:\/\/arxiv.org\/abs\/2410.17725."},{"key":"e_1_2_9_25_1","volume-title":"ICML Workshop on Challenges in Representation Learning","author":"Lee D.\u2010H.","year":"2013"},{"key":"e_1_2_9_26_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-10602-1_48"},{"key":"e_1_2_9_27_1","doi-asserted-by":"publisher","DOI":"10.1109\/TITS.2023.3334616"},{"key":"e_1_2_9_28_1","doi-asserted-by":"publisher","DOI":"10.1016\/j.eswa.2023.120644"},{"key":"e_1_2_9_29_1","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2023.3299223"},{"key":"e_1_2_9_30_1","doi-asserted-by":"publisher","DOI":"10.3389\/frai.2022.737363"},{"key":"e_1_2_9_31_1","unstructured":"Microsoft\u2010COCO.2021.\u201cMicrosoft Coco: Common Objects in Context.\u201dhttps:\/\/cocodataset.org\/#detection\u2010eval."},{"key":"e_1_2_9_32_1","unstructured":"OpenAI.2024.\u201cHello GPT\u20104o.\u201d Technical Announcement and Model Specifications.https:\/\/openai.com\/index\/hello\u2010gpt\u20104o\/."},{"key":"e_1_2_9_33_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICMLA.2017.0-179"},{"key":"e_1_2_9_34_1","doi-asserted-by":"publisher","DOI":"10.1109\/TITS.2025.3542174"},{"key":"e_1_2_9_35_1","unstructured":"Peng Y. H.Li P.Wu Y.Zhang X.Sun andF.Wu.2024.\u201cD\u2010Fine: Redefine Regression Task in Detrs as Fine\u2010Grained Distribution Refinement.\u201dhttps:\/\/arxiv.org\/abs\/2410.13842."},{"key":"e_1_2_9_36_1","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2020.3019967"},{"key":"e_1_2_9_37_1","unstructured":"Reid M. N.Savinov D.Teplyashin et\u00a0al.2024.\u201cGemini 1.5: Unlocking Multimodal Understanding Across Millions of Tokens of Context.\u201dhttps:\/\/arxiv.org\/abs\/2403.05530."},{"key":"e_1_2_9_38_1","unstructured":"Robinson I. P.Robicheaux andM.Popov.2025.\u201cRf\u2010Detr Pretrained Models.\u201d Pretrained RF\u2010DETR Model Weights (MEDIUM and LARGE).https:\/\/github.com\/roboflow\/rf\u2010detr?ref=blog.roboflow.com."},{"key":"e_1_2_9_39_1","unstructured":"Robinson I. P.Robicheaux M.Popov D.Ramanan andN.Peri.2025.\u201cRf\u2010Detr: Neural Architecture Search for Real\u2010Time Detection Transformers.\u201dhttps:\/\/arxiv.org\/abs\/2511.09554."},{"key":"e_1_2_9_40_1","first-page":"6956","volume-title":"Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition","author":"Saito K.","year":"2019"},{"key":"e_1_2_9_41_1","unstructured":"Sapkota R. R. H.Cheppally A.Sharda andM.Karkee.2026.\u201cYolo26: Key Architectural Enhancements and Performance Benchmarking for Real\u2010Time Object Detection.\u201dhttps:\/\/arxiv.org\/abs\/2509.25164."},{"key":"e_1_2_9_42_1","doi-asserted-by":"crossref","unstructured":"Schick T. J.Dwivedi\u2010Yu R.Dess\u00ec et\u00a0al.2023.\u201cToolformer: Language Models Can Teach Themselves to Use Tools.\u201d arXiv Preprint arXiv:2302.04761.https:\/\/arxiv.org\/abs\/2302.04761.","DOI":"10.52202\/075280-2997"},{"key":"e_1_2_9_43_1","volume-title":"Advances in Neural Information Processing Systems","author":"Sohn K.","year":"2020"},{"key":"e_1_2_9_44_1","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-032-08649-5_26"},{"key":"e_1_2_9_45_1","unstructured":"Tian Y. Q.Ye andD.Doermann.2025.\u201cYolov12: Attention\u2010Centric Real\u2010Time Object Detectors.\u201dhttps:\/\/arxiv.org\/abs\/2502.12524."},{"key":"e_1_2_9_46_1","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52729.2023.01392"},{"key":"e_1_2_9_47_1","doi-asserted-by":"publisher","DOI":"10.1109\/TITS.2023.3258480"},{"key":"e_1_2_9_48_1","unstructured":"Wei X. X.Cui andN.Cheng.2023.\u201cZero\u2010Shot Information Extraction via Chatting With ChatGPT.\u201d arXiv Preprint arXiv:2302.10205.https:\/\/arxiv.org\/abs\/2302.10205."},{"key":"e_1_2_9_49_1","doi-asserted-by":"crossref","unstructured":"Widjojo D. E.Setyati andY.Kristian.2022a.\u201cIntegrated Deep Learning System for Car Damage Detection and Classification Using Deep Transfer Learning.\u201d 21\u201326.","DOI":"10.1109\/ITIS57155.2022.10010292"},{"key":"e_1_2_9_50_1","doi-asserted-by":"publisher","DOI":"10.1109\/ITIS57155.2022.10010292"},{"key":"e_1_2_9_51_1","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.00987"},{"key":"e_1_2_9_52_1","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2020.2964055"}],"container-title":["Expert Systems"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/pdf\/10.1111\/exsy.70340","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/full-xml\/10.1111\/exsy.70340","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/pdf\/10.1111\/exsy.70340","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,7,16]],"date-time":"2026-07-16T09:34:06Z","timestamp":1784194446000},"score":1,"resource":{"primary":{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/10.1111\/exsy.70340"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,6]]},"references-count":51,"journal-issue":{"issue":"8","published-print":{"date-parts":[[2026,8]]}},"alternative-id":["10.1111\/exsy.70340"],"URL":"https:\/\/doi.org\/10.1111\/exsy.70340","archive":["Portico"],"relation":{},"ISSN":["0266-4720","1468-0394"],"issn-type":[{"value":"0266-4720","type":"print"},{"value":"1468-0394","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,7,6]]},"assertion":[{"value":"2026-03-03","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-06-12","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-07-06","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}],"article-number":"e70340"}}