Selected work

Representative Papers

A Hybrid Vision-Language Architecture for Automated Defect Reasoning and Report Generation in Industrial Inspection

May 26, 2026

This work addresses the longstanding disconnect between defect localization and structured reporting in industrial inspection, which has traditionally relied on manual intervention. The authors propose a decoupled three-stage pipeline: the Eyes module leverages YOLOv8-x-obb for high-precision oriented defect detection; the Bridge module maps detection outputs to structured prompts via parameter-free spatial encoding; and the Brain module employs a 4-bit quantized Qwen-2.5-1.5B model, fine-tuned with QLoRA and retrieval-augmented fine-tuning (RAFT), to generate standardized JSON reports. Evaluated on a small-scale synthetic dataset, the approach significantly outperforms generic end-to-end large models, achieving a BLEU-4 score of 0.41, a hallucination rate of only 4%, an expert rating of 8.6/10, and an inference speed of 47 tokens per second on a single T4 GPU—surpassing a 671B-parameter API baseline in overall performance.

0 citationsRead paper

IndoBERT-Sentiment: Context-Conditioned Sentiment Classification for Indonesian Text

Apr 08, 2026

This study addresses the challenge of context-dependent sentiment misclassification in Indonesian text, a limitation commonly observed in existing models that disregard topical context. To overcome this, the authors propose a joint modeling approach that integrates topical context with textual content, marking the first successful application of context-conditioned modeling to Indonesian sentiment analysis. Built upon the IndoBERT Large architecture (335 million parameters), the model is trained on 31,360 context–text pairs spanning 188 distinct topics. Evaluated on a held-out test set, it achieves an accuracy of 88.1% and a macro F1-score of 0.856, representing a substantial improvement of 35.6 F1 points over the strongest baseline. This advancement demonstrates the critical role of contextual information in enhancing sentiment classification performance for Indonesian.

0 citationsRead paper

IndoBERT-Relevancy: A Context-Conditioned Relevancy Classifier for Indonesian Text

Mar 27, 2026

This work addresses the lack of effective models for Indonesian relevance classification by proposing a context-conditioned relevance classifier built upon IndoBERT Large (335 million parameters). Leveraging an iterative, failure-driven data construction methodology, the authors curate a novel dataset comprising 31,360 annotated sentence pairs spanning 188 topics, which integrates both authentic multi-source data and targeted synthetic examples to encompass both formal and informal Indonesian text. The resulting model achieves 96.5% accuracy and an F1 score of 0.948 on the test set, demonstrating substantially improved robustness. The model and dataset have been publicly released on the Hugging Face platform to support further research in Indonesian natural language processing.

0 citationsRead paper

SocialX: A Modular Platform for Multi-Source Big Data Research in Indonesia

Mar 27, 2026

This study addresses the persistent challenge of data fragmentation in Indonesian multi-source big data research, where heterogeneous data formats, access protocols, and noise characteristics across platforms often necessitate redundant development of collection and analysis pipelines. To overcome this, the authors propose a modular, source-agnostic unified processing platform featuring a three-layer decoupled architecture—comprising data ingestion, language-aware preprocessing, and pluggable analytics—alongside a lightweight task orchestration mechanism. This design enables seamless integration of new data sources, preprocessing methods, or analytical tools. The platform, publicly available at https://www.socialx.id, has been validated through representative workflows, demonstrating a significant reduction in the barrier to entry for multi-source big data research while enhancing scalability and reusability.

0 citationsRead paper

Accelerating Large-Scale Cheminformatics Using a Byte-Offset Indexing Architecture for Terabyte-Scale Data Integration

Jan 26, 2026

This study addresses the challenges of inefficiency, hash collisions, and poor scalability in large-scale chemical database integration. The authors propose an efficient integration architecture based on byte-offset indexing that leverages full InChI strings—instead of collision-prone InChIKeys—to guarantee molecular uniqueness and data integrity. By replacing InChIKeys with complete InChI representations, the method reduces integration complexity from O(N×M) to O(N+M), enabling a high-performance data pipeline. The approach successfully integrates PubChem, ChEMBL, and eMolecules in just 3.2 hours—740 times faster than conventional methods—and extracts 435,413 validated compounds. Notably, this work also uncovers, for the first time, the occurrence of InChIKey hash collisions at the hundred-million-compound scale, highlighting a critical limitation of current standard identifiers in ultra-large chemical datasets.

0 citationsRead paper

Recent publications

Latest Papers

A Hybrid Vision-Language Architecture for Automated Defect Reasoning and Report Generation in Industrial Inspection

May 26, 2026

This work addresses the longstanding disconnect between defect localization and structured reporting in industrial inspection, which has traditionally relied on manual intervention. The authors propose a decoupled three-stage pipeline: the Eyes module leverages YOLOv8-x-obb for high-precision oriented defect detection; the Bridge module maps detection outputs to structured prompts via parameter-free spatial encoding; and the Brain module employs a 4-bit quantized Qwen-2.5-1.5B model, fine-tuned with QLoRA and retrieval-augmented fine-tuning (RAFT), to generate standardized JSON reports. Evaluated on a small-scale synthetic dataset, the approach significantly outperforms generic end-to-end large models, achieving a BLEU-4 score of 0.41, a hallucination rate of only 4%, an expert rating of 8.6/10, and an inference speed of 47 tokens per second on a single T4 GPU—surpassing a 671B-parameter API baseline in overall performance.

0 citationsRead paper

IndoBERT-Sentiment: Context-Conditioned Sentiment Classification for Indonesian Text

Apr 08, 2026

This study addresses the challenge of context-dependent sentiment misclassification in Indonesian text, a limitation commonly observed in existing models that disregard topical context. To overcome this, the authors propose a joint modeling approach that integrates topical context with textual content, marking the first successful application of context-conditioned modeling to Indonesian sentiment analysis. Built upon the IndoBERT Large architecture (335 million parameters), the model is trained on 31,360 context–text pairs spanning 188 distinct topics. Evaluated on a held-out test set, it achieves an accuracy of 88.1% and a macro F1-score of 0.856, representing a substantial improvement of 35.6 F1 points over the strongest baseline. This advancement demonstrates the critical role of contextual information in enhancing sentiment classification performance for Indonesian.

0 citationsRead paper

IndoBERT-Relevancy: A Context-Conditioned Relevancy Classifier for Indonesian Text

Mar 27, 2026

This work addresses the lack of effective models for Indonesian relevance classification by proposing a context-conditioned relevance classifier built upon IndoBERT Large (335 million parameters). Leveraging an iterative, failure-driven data construction methodology, the authors curate a novel dataset comprising 31,360 annotated sentence pairs spanning 188 topics, which integrates both authentic multi-source data and targeted synthetic examples to encompass both formal and informal Indonesian text. The resulting model achieves 96.5% accuracy and an F1 score of 0.948 on the test set, demonstrating substantially improved robustness. The model and dataset have been publicly released on the Hugging Face platform to support further research in Indonesian natural language processing.

0 citationsRead paper

SocialX: A Modular Platform for Multi-Source Big Data Research in Indonesia

Mar 27, 2026

This study addresses the persistent challenge of data fragmentation in Indonesian multi-source big data research, where heterogeneous data formats, access protocols, and noise characteristics across platforms often necessitate redundant development of collection and analysis pipelines. To overcome this, the authors propose a modular, source-agnostic unified processing platform featuring a three-layer decoupled architecture—comprising data ingestion, language-aware preprocessing, and pluggable analytics—alongside a lightweight task orchestration mechanism. This design enables seamless integration of new data sources, preprocessing methods, or analytical tools. The platform, publicly available at https://www.socialx.id, has been validated through representative workflows, demonstrating a significant reduction in the barrier to entry for multi-source big data research while enhancing scalability and reusability.

0 citationsRead paper

Accelerating Large-Scale Cheminformatics Using a Byte-Offset Indexing Architecture for Terabyte-Scale Data Integration

Jan 26, 2026

This study addresses the challenges of inefficiency, hash collisions, and poor scalability in large-scale chemical database integration. The authors propose an efficient integration architecture based on byte-offset indexing that leverages full InChI strings—instead of collision-prone InChIKeys—to guarantee molecular uniqueness and data integrity. By replacing InChIKeys with complete InChI representations, the method reduces integration complexity from O(N×M) to O(N+M), enabling a high-performance data pipeline. The approach successfully integrates PubChem, ChEMBL, and eMolecules in just 3.2 hours—740 times faster than conventional methods—and extracts 435,413 validated compounds. Notably, this work also uncovers, for the first time, the occurrence of InChIKey hash collisions at the hundred-million-compound scale, highlighting a critical limitation of current standard identifiers in ultra-large chemical datasets.

0 citationsRead paper