Integration Engineer, Clinical Data Path

About the role

You will own the end-to-end clinical data path: how studies come into our system from hospitals, how results are returned, and how we demonstrate reliability to hospital IT teams. This includes on-premise and cloud ingest, de-identification, tenancy, result delivery, and documentation for security reviews. The role involves hardening the path, including bringing gateway repositories under proper continuous integration.

Responsibilities

• On-premise ingest: Maintain a pynetdicom C-STORE service that assembles studies, handles incomplete senders, applies backpressure, preserves compressed transfer syntaxes, and spools to disk to prevent data loss.
• Cloud ingest: Develop a DICOMweb STOW-RS receiver with a streaming, bounded-memory multipart parser, per-part study-UID checks, and exactly one canonicalization and inference per request.
• Tenancy and privacy: Determine customer ownership of studies without silent defaults, apply DICOM PS3.15 confidentiality rules, enforce fail-closed checks, and manage encrypted re-identification stores under customer-held keys.
• Outbound lanes: Implement JSON callbacks, dictation-system autotext, and HL7 v2 ORU^R01 over MLLP, tailored to customer interface engines.
• Visibility: Establish per-study tracing, three-component health checks for integrations, drift detection between deployed and expected systems, and continuous integration for gateway repositories.
• Security reviews: Provide precise, honest documentation on transport, retention, and incident handling for hospital IT reviewers.
• Debugging: Investigate structural issues like unbounded memory reads or misapplied firewall rules that manifest as obscure errors (e.g., 502s or 403s), using packet captures, memory profiles, and protocol specs.

Requirements

• Experience implementing a protocol directly from its specification (e.g., DICOM, HL7), not just using a library.
• Experience running software in environments where direct access (e.g., SSH) is restricted, relying on logs or external feedback for debugging.
• Instinct to fail closed when handling patient data and ability to write precise, trustworthy documentation for hospital IT reviewers.
• Approach to debugging: prioritize packet captures, memory profiles, or protocol specs over additional log lines when errors lack clear application logs.

Skills

Helpful but not required:

• Experience with DICOM, PACS, VNA, or teleradiology systems.
• Familiarity with HL7 interface engines (e.g., Mirth, Rhapsody, Corepoint) or having built one.
• Contributions to open-source medical imaging projects (e.g., Orthanc, dcm4che, DCMTK, pynetdicom).
• Background in networking, embedded systems, or payments where wire formats were critical.
• Experience in healthcare or other regulated environments.
• Expertise in de-identification, privacy engineering, or confidential computing/hardware attestation.

Stack

• Languages: Python
• Frameworks: FastAPI, pytest
• Libraries: pynetdicom, pydicom
• Protocols: DICOM PS3.15, PS3.18, DICOMweb STOW-RS, HL7 v2 over MLLP
• Infrastructure: AWS (S3, SQS, Cognito, ECS/Fargate, WAF, ALB, CloudWatch, Secrets Manager) managed with Terraform
• Other: Docker, shell scripting for locked-down hospital environments

Logistics

This is a customer-facing role. You will:

• Join hospital IT and clinical security reviews.
• Be reachable during the customer’s clinical day.
• Travel occasionally for on-premise installations.

Location is tied to customer base: US-based for US customers, regional for international deployments. Your location is disclosed as a sub-processor in customer agreements, making it a hard constraint.

Pay

US: $140,000 to $180,000 base salary, plus an approximately 10% discretionary bonus. Equity may be offered to top candidates.

International (remote): Cash-weighted base salary set by local market rates for the country of work. Specific range shared early in the process. Offers are cash-only, with no equity.

Team

a2z was co-founded by Pranav Rajpurkar, an Associate Professor at Harvard Medical School with over 150 publications. The team includes engineers trained at MIT and Stanford, and fellowship-trained radiologists who read alongside the model daily.