Scaling Private Intelligence
DESIDOC Journal of Library & Information Technology · 13.11.2026 · 46 (6) · 701-709
DOI: 10.14429/djlit.22376 ↗Kurz gefasst
Fachartikel in der Zeitschrift DESIDOC Journal of Library & Information Technology, online erschienen am 13.11.2026. Verfasst von Rishit Arora und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 2.5).
Abstract (Originalsprache)
Defence and government bodies are increasingly adopting Artificial Intelligence (AI) for document processing, but there are severe concerns over using such cloud-based solutions. Sensitive data moves beyond institutional control, operational costs continue to escalate and dependence on external providers can compromise long-term continuity. The present study examines the feasibility of deploying AI models on in-house infrastructure as a secure and sustainable alternative to cloud services. The study tested three different AI models across five hardware tiers, from industrial-grade data-center GPUs to mobile processors, processing 1,000 scholarly articles from arXiv to extract structured metadata. The results demonstrate that industrial-grade hardware achieves 650 tokens per second for compact models, while high-end laptops reach 315 tokens per second-both sufficient for institutional-scale deployment. Desktop and mobile configurations also demonstrated sufficient performance for departmental and edge deployment scenarios. The findings demonstrate the technical feasibility of localised deployment of quantised Small Language Models (SLMs) for secure metadata extraction across diverse hardware platforms. The evaluated deployment strategy may reduce dependence on external cloud services and lower recurring operational costs for organisations with sustained document-processing workloads.
Zitieren
APA
Arora, R., & Nakashima, T. (2026). Scaling Private Intelligence. DESIDOC Journal of Library & Information Technology, 46(6), 701-709. https://doi.org/10.14429/djlit.22376
Vancouver
Arora R, Nakashima T. Scaling Private Intelligence. DESIDOC Journal of Library & Information Technology. 2026;46(6):701-709. doi:10.14429/djlit.22376
BibTeX
@article{Arora2026,
author = {Arora, Rishit and Nakashima, Tomoharu},
title = {Scaling Private Intelligence},
journal = {DESIDOC Journal of Library & Information Technology},
year = {2026},
volume = {46},
number = {6},
pages = {701--709},
doi = {10.14429/djlit.22376}
}
EUVIMED