Preprint · 2026 · Open access

Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization

Anurag Gupta

Figshare · Published 25 September 2026 · DOI 10.6084/m9.figshare.33992713

Open full textCite this paper

In brief

“Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization” is a 2026 preprint by Anurag Gupta, published in Figshare. OpenAlex counts 0 citations as of 19 November 2026. It is open access under a CC BY licence: the full text is free to read at Figshare.

Citations
0
Authors
1

Abstract

The integration of Large Language Models (LLMs) into retail pharmacy workflows presents unprecedented opportunities for reducing clinical bottlenecks, particularly in longitudinal chart review and the mitigation of alert fatigue. However, the transmission of Protected Health Information (PHI) to generalized or cloud-hosted LLMs violates strict data privacy regulations (e.g., HIPAA) and introduces severe data leakage risks. Current de-identification techniques, such as static data masking, destroy the semantic context required for accurate LLM clinical inference. In this paper, we propose a novel middleware architecture: the Bidirectional Synthetic Tokenizer. This system acts as an asynchronous proxy that intercepts raw Electronic Health Record (EHR) payloads, dynamically generating a 1-to-1 "Synthetic Shadow Profile" prior to inference. Our stateful mapping engine replaces real demographics and clinical identifiers with statistically equivalent synthetic tokens, preserving 100% of the relational clinical semantics while enforcing k-anonymity. Upon receiving the LLM's inference, the architecture reverse-maps the synthetic tokens back to the original PHI for end-user display. Experimental results demonstrate that our architecture achieves 0% PHI leakage with a semantic retention rate of 99.8%, an average inference latency overhead of 42.5 ms, and a 99th percentile (p99) latency of 58 ms. This framework enables enterprise healthcare systems to safely deploy advanced generative AI to clinicians without exposing sensitive patient data.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Gupta, A. (2026). Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization. Figshare. https://doi.org/10.6084/m9.figshare.33992713

MLA 9

Gupta, Anurag. “Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization.” Figshare, 2026. https://doi.org/10.6084/m9.figshare.33992713.

Chicago

Gupta, Anurag. 2026. “Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization.” Figshare. https://doi.org/10.6084/m9.figshare.33992713.

Harvard

Gupta, A. (2026) ‘Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization’, Figshare. doi:10.6084/m9.figshare.33992713.

IEEE

A. Gupta, “Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization,” Figshare, 2026, doi: 10.6084/m9.figshare.33992713.

BibTeX

@misc{gupta2026privacy,
  title     = {{Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization}},
  author    = {Gupta, Anurag},
  howpublished = {Figshare},
  year      = {2026},
  doi       = {10.6084/m9.figshare.33992713},
  url       = {https://doi.org/10.6084/m9.figshare.33992713}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data

  • k-anonymity
  • de-identification
  • protected health information
  • electronic health records
  • alarm fatigue
  • pharmacy workflow
  • HIPAA compliance
  • data loss prevention

Work with this paper

Most-cited papers in Computer Science

All highly cited Computer Science papers

Questions about this paper

How do I cite this paper in APA style?

Gupta, A. (2026). Privacy-Preserving Clinical LLM Inference via Bidirectional Synthetic Tokenization. Figshare. https://doi.org/10.6084/m9.figshare.33992713

How many times has this paper been cited?

0 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access at Figshare: sign in and open the full text from this page.

Who wrote this paper?

Anurag Gupta.

What is the DOI of this paper?

10.6084/m9.figshare.33992713, permanently at https://doi.org/10.6084/m9.figshare.33992713. Sign in and open it from this page.

Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.