Skip to content
Just PrivacyDottie™

A little genius. A lot of discretion.

Big AI.
Smaller data
footprint.

Meet Dottie. A proprietary privacy layer designed to detect, redact, and obfuscate patient identifiers before clinical information reaches external AI models.

See Dottie at workMeet your privacy layer

Clinical context, with less identifying detail.

Dottie, an illustrated privacy expert with a chin-length bob and a white lab coat, adjusting her gold glasses.

IDENTITY: SENSITIVE

Let’s keep that private.

CONTEXT: VALUABLE

Keep the clinical picture.

Dottie. Your resident privacy nerd.

01 /

The intelligence behind the discretion

The models move fast.
Privacy should
keep up.

Healthcare AI vendors build on enterprise foundation models from companies such as OpenAI and Anthropic to benefit from rapidly advancing capabilities. That makes the flow of clinical information—and the safeguards around it—essential.

BAAs, DUAs where applicable, HIPAA/HITECH policies and procedures, and enterprise model terms establish important protections. Healthcare AI also calls for practical controls that reduce the identifiable information shared with external services.

Dottie adds a proprietary layer of PHI and PII redaction and obfuscation within that workflow. The goal: minimize exposure, preserve useful clinical context, and reduce the downstream risk of inappropriate retention, disclosure, or use.

A smarter point of protection. Before transmission.

01 / The starting point

Clinical source

Clinical information begins within the health system’s workflow.

02 / The privacy layer

Dottie gets to work.

Detect, redact, and obfuscate identifying details before external AI processing.

03 / The next step

AI tool

Useful clinical context moves forward with less identifiable information.

02 /

Privacy, made tangible

She sees the details.
Then minds them.

Patient identifiers aren’t the clinical story. Explore a simple example of how redaction can reduce identifying detail while keeping the clinical context useful.

LESS IDENTITY. MORE CLINICAL CONTEXT.

CLINICAL EXCERPTIllustrative example · synthetic data

Redaction mode

Patient identifiers out. Care team, facility and city stay for context.

Patient
Maya Bennett
DOB
04/12/1973
MRN
0049182
Phone
202-555-0147
Physician
Dr. Elena Park
Facility
Riverside Cancer Center
City
Arlington, VA
Diagnosis
Early-stage breast cancer
Receptors
ER+ / HER2−
Plan
Partial breast irradiation
Clinical context retained. Direct identifiers redacted in this example.

Visual demonstration only; this page does not process clinical data. This excerpt illustrates selected identifiers, not a complete de-identification assessment.

03 /

Zero retention. By design.

Nothing kept.
Nothing to
configure.

Dottie is our flagship zero-retention model. Clinical information is processed in memory for the moment it takes to protect it, then released. There is no storage setting to switch on, because storage was never built in. The key that restores identities is encrypted and held by you, or kept only in your browser’s memory, never by Dottie.

Data retention

0Zero is the only setting.

What Dottie doesn’t have

  • A storage toggle
  • Retention tiers
  • PHI in logs
  • A copy of your data
  • 01

    Processed in memory

    Inputs are handled transiently to detect, redact, and obfuscate, then discarded.

  • 02

    No storage options

    No retention tiers, no archive toggle, no opt-in history. There is nothing to misconfigure.

  • 03

    Never used for training

    Clinical content is not used to train or improve Dottie or any downstream model.

  • 04

    Less to protect later

    What isn’t kept can’t be disclosed, breached, or repurposed downstream.

04 /

Local or dedicated. Your call.

Two ways to run Dottie.
Same protection.
Your choice of where.

Run Dottie on the device in front of you, or on a dedicated server built for clinical data. Either way, identifiers come out before your AI sees the text and go back in when the work comes home.

Option Adottie-mini

In your browser

A compact model that runs on the user’s own GPU through WebGPU.

  • Identifiable text never leaves the browser tab. Only redacted text goes onward.
  • Detection, redaction, the leak check and re-injection all run locally.
  • The map that restores identities lives in memory for the session, and is never sent or stored.
  • A one-time model download of about 350 MB, cached by the browser.

If a device can’t finish in time, Dottie can route the work to the dedicated server instead.

Option Bdottie-max

Dedicated secure server

Our larger model on a dedicated GPU server in a BAA-covered Google Cloud environment.

  • HTTPS only, with keyed access for every request.
  • No third-party AI or OCR service sees the raw text. Scanned pages are read on the same server.
  • The re-identification vault is encrypted, bound to your organization and case, and returned to your system.
  • Dottie’s audit log records counts and outcomes, not clinical text.

The most accurate option, and the one that handles scanned documents.

The round trip

  1. 01

    Detect

    Rules and the model work together to find identifiers across the documents.

  2. 02

    Redact

    Identifiers become tokens or stand-ins that keep their shape.

    Mr. [PATIENT_LAST_1]
  3. 03

    Your AI works

    Dates stay on the timeline, so delays and deadlines still add up.

    [DATE_2 day+14]
  4. 04

    Verify

    The output is checked for any identifier the AI repeated.

  5. 05

    Re-inject

    Real values go back exactly, including dates the AI wrote itself.

If Dottie can’t safely remove an identifier, nothing is sent.

05 /

Measured, not promised

The numbers.

We test Dottie end to end: identifiers out before the AI, the right patient back in the finished document. Here is where the November 2026 evaluation landed.

  • 0 of 1,666

    patient identifiers reached the downstream AI

  • 0 of 60

    never-seen bundles with a missed name, DOB, MRN or member ID

  • 0 of 287

    identifiers missed for patients with accented names

  • 0 of 183

    finished documents with a stand-in name left after re-injection

  • 12 of 12

    prompt-injection attacks blocked

  • 99.8%

    identifier recall for the in-browser model

November 2026 synthetic evaluation with no real patient data. These results do not establish a real-world error rate.

The Just Healthcare Solutions suite

Part of the suite.
Now with Dottie.

Just Privacy joins the products health systems already use to secure access, connect systems, and move cases forward, with Dottie as its flagship privacy layer.

Explore Just Healthcare Solutions

Serious about privacy. Delightfully Dottie.

Welcome to the future
of data privacy protection.

Let’s talk Dottiehello@justhealthcaresolutions.com