Active Crossing is a healthcare integration practice built on fifteen years of HL7 v2, C-CDA and FHIR R4 delivery. We design, build and operate the interfaces that connect health systems, HIEs and vendors — and we bring AI into that work without letting protected health information leave your network.
Health organizations want to use large language models on clinical data, but that data cannot leave the network with protected health information intact. Existing de-identification tools were built for free-text notes — they do not handle the formats that actually carry clinical data between systems, where PHI sits in structured fields and in embedded narrative. We are building the engine that closes that gap, and closes the loop back to the patient.
HL7 v2, C-CDA and FHIR R4 are parsed as structured data, not flattened to text. Identifiers are replaced with random surrogates; dates shift consistently per patient so longitudinal records stay analyzable.
Your networkThe link between each real patient and their surrogates is held in a local mapping vault. It never leaves your infrastructure, and the output cannot be reversed without it.
Your networkDe-identified payloads can safely reach a hosted frontier model for summarization, problem-list extraction or timeline questions — the capable models you could not otherwise use.
Hosted LLMResults come back and are mapped to the right patient on-premises, so the output is usable in care rather than stranded in an anonymous dataset.
Your networkSchema-driven rules resolve identifiers in structured fields, including the custom Z-segments, misused fields and repeating components that real-world messages are full of.
A locally hosted language model handles PHI in narrative sections and free-text values — OBX-5, NTE segments, C-CDA narrative blocks. It sees only the fields rules cannot resolve, with their structural context, which keeps the system fast and auditable on a single on-premises server.
The same patient appears in HL7 v2, C-CDA and FHIR under different field layouts and identifier types. Cross-format identity resolution links those records before substitution so surrogates stay consistent everywhere.
Removing PHI is only half the problem. We quantify how much downstream task performance survives de-identification, comparing results on original versus de-identified input — evidence privacy officers and data governance committees can act on.
Status: early prototype, under active development. All current work uses synthetic and public de-identified data. If you run an HIE or a health-system analytics team facing this problem, we are looking for design partners — get in touch.
Engagements are led personally by a certified integration architect who has served as primary technical owner on more than thirty implementations — from first discovery call through architecture, build, go-live and ongoing support.
Certified Mirth Connect (2024) architecture, build and rescue work on engines processing millions of daily transactions.
Standards-grade design for exchange between health systems, HIEs, health plans and technology vendors.
LLM systems designed for clinical environments, where the privacy model is part of the architecture — not an afterthought.
Matching the same human across systems that disagree about who they are.
The infrastructure around the engine: observability, pipelines and the automation that keeps delivery repeatable.
Technical leadership for teams that need an architect in the room, not another ticket queue.
Most healthcare AI projects stall at the same place: the data that would make the model useful is the data legal will not let you send anywhere. We build systems that resolve that tension in the architecture.
Our own de-identification engine is the product this thinking comes from — a closed on-premises loop that lets a hosted model work on clinical data without ever seeing PHI.
Senior Integration Architect with more than fifteen years in healthcare interoperability. Ryan designs and operates the HL7 v2, C-CDA and FHIR integrations that move clinical data between health information exchanges, providers and health-IT vendors — as enterprise architect inside EHR platform teams, and across consulting engagements where he served as the primary technical owner from discovery to go-live.
His recent work centers on the AI-driven automation of integration delivery itself: automated discovery, MCP connectors and generated interfaces, backed by PR-gated CI/CD and Python diagnostics. Earlier in his career he led development on large-scale data systems in logistics and finance, including a global shipment rating platform handling international freight pricing at scale.
Integration engine rescue, a TEFCA onboarding, an AI initiative that keeps running into PHI — tell us what you're working on and we'll tell you honestly whether we're the right fit.
