AI for patient data systems

We turn patient data into clinical intelligence.

Pretty Fire is a physician-led AI venture building the systems that read, organize, and reason over the messy reality of patient data — so clinicians and their teams can act on what matters, sooner.

Since 1996
Building clinical data systems
100,000+
Records managed at HMO scale
Zero
Security breaches, trial to production
The thesis

Medicine doesn't have a data shortage. It has a data legibility problem. The record of a patient's journey is scattered across notes, messages, images, and systems that never talk to each other — and the intelligence is lost in the gaps.

Pretty Fire closes those gaps. We build AI that assembles a coherent, voice-attributed story of each patient and surfaces the next right action — without ever compromising privacy.

Architecture

A system that learns without forgetting

Navigator is built on a layered intelligence model. New information appends; only what's new gets analyzed; profiles re-synthesize incrementally. Nothing is ever reprocessed from zero.

1 · Raw voice log

An append-only, chronological record of every voice-attributed input. Never modified, never re-analyzed.

2 · Extracted signals

Structured data drawn from each input — incrementally, so only new entries trigger work.

3 · Synthesized profiles

Clinical, financial, and engagement profiles that re-synthesize only when new signals arrive.

4 · Guidance

Ephemeral, on-demand recommendations generated from the current state — the next right action.

Our story

We were securing patient data before there was a HIPAA

Pretty Fire didn't start with a model. It started with a software company in the 1990s and a stubborn belief that the same engineering discipline could make patient data safe, structured, and genuinely useful. Three decades later, that belief is the foundation everything else is built on.

1990s · PHAEDRUS INC. · NEWTON, MA

It began with the web

Founded while its co-founder was still in medical school at Yale, Phaedrus Inc. was a software firm that built websites for marquee clients — the FleetCenter (home of the Boston Bruins and Celtics), the Bruins themselves, and the USS Constitution site for the U.S. Navy. By 1998 it had already received and turned down buyout offers. The instinct was set: build systems that big institutions can trust.

1996 · YALE UNIVERSITY

CLAY is born

Phaedrus turned its engineering from the web to medicine. CLAY — the Collaborative Layered Application — first ran at Yale as a platform to coordinate research on plastic surgery patients. It became the subject of a published research paper, and the start of a long bet: that clinical images and the data around them belong in one secure, structured record.

2005 · mdconsult.net / PHAEDRUS

HIPAA-compliant visible-light imaging

CLAY v1.0 was one of the first systems ever built to capture, store, and share clinical visible-light imaging — dermatology, wound care, pathology, plastic surgery before-and-afters — under HIPAA. Developed in consultation with the U.S. Department of Health and Human Services before the Privacy and Security Rules were finalized: encrypted access, full audit logging, hourly redundant backup, and DICOM integration with legacy hospital systems.

ENTERPRISE SCALE

Proven at one of the nation's largest HMOs

CLAY went on to manage 100,000+ image-rich medical records at HMO scale — turning loose photos on memory cards into a compliant, searchable record. Across clinical trial and production: zero unplanned downtime, and not a single security breach.

TODAY · PRETTY FIRE

AI that reasons over the whole journey

Pretty Fire brings modern AI to that foundation. Navigator is the result: a system that reads the full, multi-voice record of care and turns it into intelligence a team can act on — privacy-first, physician-led, and built where the work actually happens.

What we won't compromise

Principles

01

Privacy is the architecture

Identifiers live in a separate vault. AI reasons over de-identified data. Privacy isn't a setting we add later — it's the shape of the system.

02

Physician-led, not physician-replacing

Built by and for clinicians. The system surfaces signal and proposes action; the human stays in command of the decision.

03

Patient safety first

Every output is conservative by default. We would rather under-claim than overstate — in a clinic, a confident wrong answer is the dangerous one.

04

Real deployment, not demos

We build systems that survive contact with messy, real-world data and daily clinical use — not just the happy path.

05

Earn trust incrementally

The system learns without forgetting and improves a little every day. Trust compounds the same way.

06

Useful beats impressive

If it doesn't make the next action clearer for a real team, it doesn't ship. Utility is the only metric that counts.

Let's make patient data legible.

Whether you're a clinic, a health-tech team, or an investor in responsible AI for medicine — we'd like to hear from you.