Clinical risk & crisis
Suicide and safety risk assessment, ED psychiatric evaluation, disposition decisions, safety planning, and crisis intervention under incomplete information.
Clinical Product Safety · Behavioral Health AI Evaluation
Clinical judgment translated into safer AI systems.
Licensed Clinical Social Worker with 20+ years in behavioral health, combining crisis and risk assessment with hands-on AI evaluation, guardrail design, red-teaming, escalation logic, and independent human verification.
20+ yearsBehavioral health experience
3 yearsEmergency psychiatric evaluation
9Clinical AI risk categories
34Gold-standard acceptance cases
8Negative controls for over-blocking
The clinical → AI bridge
High-consequence judgment, made legible to product and engineering teams.
My work focuses on identifying harm, distinguishing missing evidence from acceptable uncertainty, assigning escalation and ownership, and testing systems for both missed risk and unnecessary blocking.
AI safety & evaluation
Red-teaming, gold-standard case design, unsafe-output criteria, risk taxonomies, guardrails, and missed-risk versus false-positive tradeoffs.
Clinical-to-technical translation
Turning clinical judgment into severity rules, escalation pathways, evidence requirements, review boundaries, and requirements teams can implement.
Selected clinical AI safety work
Guardrails, evaluation, and human authority.
I built an independent Clinical AI Safety Lab to test a practical question: how can AI support behavioral-health work without making clinical, privacy, or authority decisions it should not make?
From instruction-only safeguards to a guarded workflow
- Red-teamed crisis timing, safety-clearance, PHI, and diagnosis requests.
- Documented failures across two versions.
- Rebuilt the workflow with pre-model and post-model checks, eight generic-only task modes, and fail-closed ambiguity handling.
Measurable risk, routing, and acceptance criteria
- Built a nine-category clinical AI risk taxonomy with defined routing and redirect behavior.
- Wrote a 34-case gold-standard acceptance suite.
- Included eight negative controls to measure over-blocking as well as missed risk.
Human approval before ingestion
- Built a clinical reference-library pipeline with automated content-safety screening.
- Added five-category triage labels and required human approval before ingestion.
- Used refusal testing to surface a guard miss and route it to targeted repair.
Independent review instead of inherited verdicts
- Defined unsafe-output criteria for crisis, diagnosis, treatment, protected details, and safety-clearance language.
- Designed independent verification from primary evidence.
- Flagged reviewer role conflicts rather than allowing silent authority overlap.
Clinical judgmentRisk, context, authority
Risk taxonomyNine categories and routing
Input checkScope, mode, PHI
Model / RAGBounded task
Output checkUnsafe criteria
Human reviewApprove, block, repair
Experience
Clinical depth with a long technology thread.
My background spans direct behavioral-health care, emergency psychiatric consultation, supervision, quality review, and earlier technology support and distance-learning work.
Penobscot Community Health Center · Clinical Therapist
Psychosocial assessment, therapy, treatment planning, crisis intervention, multidisciplinary support, critical-incident follow-up, chart review, and supervision.
Acadia Hospital Consult Services · Emergency Psychiatric Consultation Clinician
Emergency psychiatric evaluations, suicide and safety risk assessment, disposition recommendations, safety planning, and statewide placement coordination.
Additional clinical leadership
Behavioral therapy, medical social work, program supervision, outpatient substance-use treatment, and inpatient clinical work.
University technology roles
Online-class development and technical support, satellite-office technology deployment, help desk operations, residence-hall networking, and translating between users and technical staff.
Documents
Résumé and selected AI safety portfolio.
The documents below are the concise evidence set I use alongside applications for clinical AI safety, product safety, and behavioral-health technology roles.
Résumé
Clinical product safety, behavioral health AI evaluation, crisis and risk assessment, clinical leadership, and technical environment.
Download résumé PDFClinical AI Safety Portfolio
One-page evidence summary covering the guarded assistant, risk taxonomy, test suite, reference-library safety, and independent verification.
Download portfolio PDF
Contact
Clinical expertise for AI products where safety matters.
Based in Bangor, Maine. Licensed Clinical Social Worker in Maine. Interested in clinical AI safety, behavioral-health technology, AI evaluation, product safety, and clinical-to-technical product roles.