EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Edge AI for On-Site Health Risk Scoring: A RAG-Enabled Framework

Rama Krishna Kumar Lingamgunta, Abhijit Ubale, Siva Kantha Rao Vanama

American Journal of Technology · 2025

DOI: 10.58425/ajt.v4i3.451 ↗

Kurz gefasst

Fachartikel in der Zeitschrift American Journal of Technology, 2025. Verfasst von Rama Krishna Kumar Lingamgunta und 2 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Aim

This study aims to develop a privacy-preserving, on-device framework for health risk scoring during in-home insurance care assessments. The goal is to enable field practitioners to perform real-time clinical evaluation, coding, and documentation without dependence on cloud connectivity or external computation.

Methods

A Retrieval-Augmented Generation (RAG) framework was implemented on an Android-based edge environment. Structured and unstructured visit data were locally processed using lightweight small language models (SLMs) and encrypted vector indexes containing payer rules, HCC mappings, and CMS guidelines. A simulated dataset was generated using SYDA (an open-source synthetic data platform) to replicate diverse member profiles and evaluate inference accuracy, latency, and compliance alignment.

Results

The system successfully generated on-site risk scores mapped to ICD-10 and HCC categories without transferring protected health information (PHI). Experimental evaluations demonstrated a ~37% reduction in decision latency and a 20% improvement in documentation accuracy compared to cloud-based methods. Compliance verification with CMS-HCC standards was achieved instantly due to local rule retrieval.

Conclusion

This study demonstrates that integrating Edge AI with local RAG architectures can transform insurance care management workflows by delivering faster, auditable, and privacy-compliant health risk assessments at the point of care. The findings show that payer-side analytics can achieve both interpretability and scalability through edge-based computation. Recommendations: This study recommend future research to focus on integrating HL7 FHIR APIs for real-world EHR synchronization, extend the framework to multimodal health data (speech and imaging), and apply federated edge - cloud learning for adaptive model refinement.

Zitieren

APA

Lingamgunta, R. K. K., Ubale, A., & Vanama, S. K. R. (2025). Edge AI for On-Site Health Risk Scoring: A RAG-Enabled Framework. American Journal of Technology. https://doi.org/10.58425/ajt.v4i3.451

Vancouver

Lingamgunta RKK, Ubale A, Vanama SKR. Edge AI for On-Site Health Risk Scoring: A RAG-Enabled Framework. American Journal of Technology. 2025. doi:10.58425/ajt.v4i3.451

BibTeX
@article{Lingamgunta2025,
  author = {Lingamgunta, Rama Krishna Kumar and Ubale, Abhijit and Vanama, Siva Kantha Rao},
  title = {Edge AI for On-Site Health Risk Scoring: A RAG-Enabled Framework},
  journal = {American Journal of Technology},
  year = {2025},
  doi = {10.58425/ajt.v4i3.451}
}
Kostenlos starten →