Clinical AI assistant for chat-based telemedicine
28/02/2026 às 09:12Descrição do Projeto:
Context
We are a telemedicine provider running chat-based clinical appointments (including mental health/psychiatry). We want to build a clinician-assist AI workflow powered by the OpenAI API to improve speed, consistency, and documentation quality — while preserving physician oversight and strong privacy controls.
1) Objective
Build an AI assistant that supports clinicians during chat appointments by:
• collecting structured information (intake / triage / anamnesis),
• producing clinician-facing summaries,
• drafting patient-facing explanations and follow-up instructions,
• drafting a structured clinical note / medical record,
• routing emergencies / red flags appropriately.
Key principle: This is decision-support, not autonomous care. The clinician must explicitly approve critical steps and any patient-facing medical guidance.
2) Non-Negotiable Requirements
A) Human-in-the-loop “stage gating”
Implement the appointment workflow as a state machine with explicit clinician approval gates:
• The assistant must never advance stages automatically.
• Only clinician actions (button click / clinician-only command) move the workflow forward.
• Patient messages must never authorize progression.
B) 3-phase workflow (high level)
We use a three-phase structure (details shared after NDA):
• Triage support — generate a small set of targeted follow-up questions
• Clinical analysis + draft plan — clinician-facing assessment + patient-friendly draft guidance
• Medical record draft — objective, technical documentation (EHR-style)
C) Safety & escalation
Implement “hard-stop” escalation logic for red flags (examples: suicidality/self-harm intent, severe acute symptoms). The system must interrupt the normal flow and route to urgent guidance + clinician escalation.
D) Privacy-first development (no PHI exposure)
We want the system designed so that:
• No real patient PHI is used during development/testing.
• Testing uses synthetic and/or properly de-identified data only.
• Implement a PHI/PII minimization & redaction layer before any OpenAI API call.
• Provide guidance for best-available OpenAI API privacy posture (e.g., retention-minimization options depending on eligibility).
3) Technical Expectations
Must be able to:
• Integrate with the OpenAI API and enforce structured JSON outputs (schema validation)
• Build workflow orchestration (state machine, transitions, retries, fallbacks)
• Implement prompt-injection resistance patterns and hallucination containment
• Provide auditability (who approved what, when) without leaking PHI
• Build a simple clinician review UI (or integrate with an existing dashboard) for approve/edit/reject/escalate
Nice to have:
• Experience with WhatsApp/chat systems and webhook-based integrations
• Healthcare security mindset (LGPD/HIPAA-like design patterns)
4) MVP Deliverables
• Architecture + security/privacy notes
• State machine implementation (3 phases + gating)
• JSON schemas per phase + validation
• PHI redaction/minimization gateway
• Working MVP demonstrating:
• triage question generation
• clinician approval gate
• escalation flow for red flags
• clinician-facing summary + patient-facing draft text (as draft, clinician-reviewed)
• medical record draft
• Synthetic test suite (including red-flag cases)
• Documentation for clinicians/operators
5) Acceptance Criteria (Objective)
• Outputs consistently pass schema validation
• Red-flag scenarios trigger escalation in 100% of test cases
• Workflow cannot progress without clinician approval
• No PHI required in dev/test environment
• Clear audit trail for approvals/edits
6) What to include in your application
Please send:
1–3 examples of shipped LLM/agent systems (preferably with integrations)
Proposed stack + workflow approach (short)
How you’ll implement privacy-first development and PHI minimization
Quote for a paid pilot (first stage + approval + escalation)
Confidential details (exact prompts, templates, phrasing standards, integration endpoints, internal policies) will be shared only with shortlisted candidates after NDA.
Atividades do cliente nesse projeto:
Última visualização: 23/03/2026 às 11:02
Última interação: 28/02/2026 às 14:52
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
-
Submetido: | Oferta: Privado | Duração estimada: Privado
Desenvolvedor de agentes de IA - projeto AI-First para grupo de turismo - Somos um dos maiores grupos de turismo do Brasil, com 80 colaboradores e quase 30 anos de mercado. Estamos iniciando uma transformação AI-First e buscamos um desenvolvedor PJ remoto p...
Desenvolvedor de IA para assistente clínico em telemedicina - Somos a Doctor Friend, uma plataforma brasileira de telemedicina. Buscamos um desenvolvedor especialista em IA para implementar um assistente clínico inteligente integrado ao nosso sistema.
Automação de qualificação de leads via WhatsApp - Estou iniciando uma operação focada em automação, inteligência artificial e crescimento para pequenas e médias empresas brasileiras. Neste primeiro momento...
Configuração do OpenClaw Manager na Hostinger (VPS/Docker/IA) - Olá! Preciso de um profissional técnico com experiência em VPS, Docker e ferramentas de IA para me auxiliar na configuração e no deploy do OpenClaw Manager na minh...
Integração Adapta One (Expert) com WhatsApp Business e Meta - Preciso configurar a integração do meu atendente de IA criado no Adapta One (Expert) para atendimento nos canais: - WhatsApp Business - Messenger - Instagram - Facebook Se você...
