Connect what matters.
Work across hospital interfaces, clinical vocabularies, data pipelines, and the systems that make information usable.
Careers & consulting opportunities
Your expertise.
Better-connected care.
Help healthcare teams make their systems work together. Bring your experience in engineering, clinical informatics, or program delivery to work that reaches the people providing care.
The work behind the technology
Bring clinical and technical work together.
Collaborate with engineering and clinical leadership, and the facility teams who know how care is delivered.
See the work through.
Help move projects from assessment to implementation, with testing, documentation, and a clear handover.
Find your area of expertise
Opportunities at CareCompile
Employee and consulting opportunities, with working arrangements listed for each role. Select a title to see the details.
1 role types
Senior Software Engineer, AI-Assisted Stewardship
Accepting applicationsView role +
Senior Software Engineer, AI-Assisted Stewardship
Rural hospitals and clinics send lab, medication, diagnosis and encounter data in dozens of local dialects. A dictionary turns all of it into one language: LOINC, SNOMED CT, RxNorm and ICD-10-CM, published as versioned FHIR artifacts the whole program runs on. Software proposes the code bindings. A credentialed terminologist decides. You build the software half of that sentence, and the record that proves it. Models run on open weights inside the program’s own environment: nothing goes to an outside service, nothing is trained on program data, and every proposal carries its evidence, its model version and its reason into a decision record an auditor can read. Base period engagement is ten months with four 12-month option years to follow, starting when the program agreement is executed, expected in the fourth quarter of 2026. Offers are made now, contingent on that date. No patient data: all work is on synthetic and de-identified data.
What you would do
- Build the stewardship service behind the terminologist’s console: the orphan queue that quarantines every unmapped local code instead of dropping it, the approval path no binding escapes, and the decision record that carries who decided, what they decided, why, and which model and prompt version proposed it. Live by month five, in daily use by month seven.
- Build the binding-proposal harness: open-weight models served inside the program environment, retrieval across the loaded vocabularies, ranked candidates with evidence a reviewer can check in seconds. First scaffold in month two, proposals flowing into the queue by month five.
- Build the evaluation harness that keeps the models honest: an evaluation set drawn from known-good mappings, acceptance rate measured by domain, and the 30-day disposition clock computed from the record rather than estimated. You will be asked whether the AI is helping, and you will answer with numbers.
- Own the conformance and regression suites: the interface with the platform team freezes on day 45, and from then on every dictionary release is proven against the one before it. Your suite is the acceptance evidence for the harmonization service, so when it passes, the milestone is delivered.
- Build the quality-metrics service and the per-facility scorecards behind the evaluation reports due at month six and again at month ten: mapping success rate by facility, source, domain and vocabulary, and orphan concepts found and resolved.
- Own the build pipeline and the record: signed, versioned artifacts with test evidence for every release, model and prompt versions pinned to each dictionary version, and the runbooks the standing stewardship team inherits at the end of the base period.
Experience you bring
- Eight or more years in software engineering, with production ownership of backend services and the pipeline that ships them. You have been on call for something people depended on.
- You have shipped an LLM-assisted feature that real users relied on, and you measured it. You know what your model gets wrong, how often, and what you changed after you found out.
- Test automation that a regulated or high-consequence organization trusted, and an opinion about what makes a passing test worth believing.
- Go, Java or Python to production standard, plus SQL, event-driven and REST service design, and observability that tells you a service is wrong before a person does.
- Comfort building inside a hard boundary: open-source components only, no proprietary runtime dependency, no program data leaving the environment and no model trained on it. The constraint is the job, not an obstacle to it.
- Prove yourself: In your note, describe one time a model you shipped was confidently wrong, how you found out, and what you changed. Nice to have: open-weight model serving (vLLM, Ollama, llama.cpp), retrieval and vector search, OIDC and role-based access, audit-log design, Kubernetes, and any exposure to HL7 v2, FHIR, laboratory systems or clinical vocabularies.
Upcoming program work emphasizes open-source infrastructure, documented decisions, and human review of AI-assisted output. Engagement duration, location eligibility, and any screening requirements will be confirmed during recruitment.
Open role page → Apply for this role →Interested in future work with us?
Not every role is listed. If your experience fits the kind of work we do, we would like to hear from you.
Start a conversation
Tell us what you bring.
Share your experience, availability, and résumé so our team can consider you for future opportunities.
Submitting interest does not guarantee an interview, offer, or project start. We will contact you if there is a potential fit.
Direct contact only. CareCompile contracts directly with individuals and with the organizations we serve. We do not work with staffing agencies, recruiters or intermediaries, and we do not answer their submissions.
Need help with the form? Contact CareCompile.
Recruiting privacy · September 11, 2026
Your information, for recruiting.
CareCompile collects the contact details, professional background, preferences, and résumé you submit to evaluate your interest and communicate with you about recruiting. If you opt in to future opportunities, we may also consider you for other roles.
Application records and résumés are available to authorized CareCompile administrators through a sign-in protected review area. Résumés are not published on the website. This form does not automatically send your information to a job board or use AI to score your application.
We retain submissions for recruiting follow-up while relevant to the opportunity or your opted-in talent profile. To request a correction, withdraw your interest, or request deletion, contact hello@carecompile.com and include your application reference. Any retention required for recordkeeping will be addressed when we process your request. See our general privacy notice.