<![CDATA[Guru Pendidikan Agama Islam (PAI) di wilayah kepulauan menghadapi tantangan dalam menyusun modul ajar yang sesuai standar Kurikulum Merdeka akibat keterbatasan pemahaman teknologi pendidikan dan tingginya beban kerja administratif. Kegiatan pengabdian kepada masyarakat ini bertujuan mendampingi guru PAI di SMPN Satu Atap 01 Pulau Pari, Kepulauan Seribu, dalam memanfaatkan media generatif berbasis Artificial Intelligence (AI) dengan pendekatan deep learning untuk menyusun dan memvalidasi modul ajar. Kegiatan dilaksanakan pada 11 Mei 2026 melalui tahapan sosialisasi, pelatihan, penerapan teknologi, serta pendampingan dan evaluasi, diikuti oleh guru SD dan SMP di Pulau Pari. Evaluasi program dilakukan melalui survei kepuasan peserta (n=25) yang mencakup dua aspek, yaitu pelaksanaan acara dan kualitas fasilitator. Hasil survei menunjukkan tingkat kepuasan pelaksanaan acara sebesar 87% (Sangat Puas) dan kepuasan terhadap fasilitator sebesar 93% (Sangat Baik), dengan skor tertinggi pada aspek pelayanan panitia dan penguasaan materi oleh fasilitator. Peserta memberikan masukan berupa penambahan contoh praktis, variasi metode, dan perpanjangan sesi diskusi. Temuan ini mengindikasikan bahwa pendekatan pendampingan yang kontekstual dan berbasis praktik efektif membangun penerimaan awal guru terhadap teknologi AI generatif, meskipun keberlanjutan perubahan praktik penyusunan modul ajar memerlukan siklus pendampingan lanjutan. Teachers in island regions face limited access to professional development amid the demands of the Kurikulum Merdeka (Independent Curriculum), which requires the independent and systematic compilation of teaching modules. This community service activity aimed to assist teachers at SMPN Satu Atap 01 Pulau Pari, Thousand Islands, in utilizing artificial intelligence (AI)-based generative media to compile Islamic Religious Education (PAI) teaching modules along with their supporting tools. The activity was carried out on May 11, 2026, through five stages: socialization, training, technology application, mentoring and evaluation, and the completion of evaluation instruments. The evaluation employed a one-group pretest-posttest design involving 24 teachers using a five-item self-efficacy questionnaire on a 1–5 scale, supplemented by data collection on module component completeness, estimated compilation time, and participant satisfaction surveys. The results showed an increase in the average perceived capability score from 2.44 to 4.68 with an N-Gain of 0.88 in the high category, presenting a statistically significant difference (Wilcoxon, p < 0.001). A total of 23 out of 24 participants experienced an increase, and none decreased. All participants produced a complete draft of a teaching module, with 23 of them containing all five complete components ranging from learning objectives to learning media and student worksheets. Furthermore, the estimated compilation time for a single module dropped from a majority taking over three hours to less than one hour. Remaining challenges included system limitations in generating visual illustrations, handling materials with mathematical symbols, and breaking down learning achievements into learning objectives. Recommendations include tiered mentoring, the provision of digital guides and online consultation channels, and expert evaluation of module quality in the next cycle.]]>
