Voice AI Space Singapore - Zixin (Zane) Yong @ OutcomesAI.

Outcomes AI is developing AI-enabled voice agents integrated with licensed nurses to address the severe healthcare staffing shortages in the United States. By automating administrative and preliminary clinical tasks, the technology aims to scale healthcare delivery safely.

Key Products

• Virtual Care: This product supports patients recovering at home by managing virtual check-ins. Patients use connected devices to monitor vital signs like blood pressure and blood sugar, while the AI voice agent conducts routine check-ups, escalating to human nurses only when necessary.
• Nurse Triage: In this clinical workflow, the AI acts as the first point of contact for inbound patient calls. It gathers demographic details, performs initial symptom assessments, and prepares a summary. To ensure clinical safety and regulatory compliance, the AI does not provide medical advice; instead, it transfers the case to a licensed nurse who makes the final clinical decision.

Technical and Operational Challenges

Building voice AI specifically for healthcare involves addressing challenges across multiple layers:

• Model Layer: The system uses a cascade pipeline of speech-to-text, large language models (LLMs), and text-to-speech. It requires specialized medical vocabulary training and robust instruction-following capabilities to handle long, complex conversations. Additionally, specialized end-of-utterance detection is needed, particularly for elderly populations who may pause frequently.
• Agent and Orchestration Layer: Managing context across long conversations often requires coordinating multiple sub-agents. Developers must balance the trade-off between latency and intelligence, as patients expect rapid responses, yet higher intelligence requires larger, slower models. Evaluating these systems across countless clinical scenarios is also highly complex and costly.
• Product and Domain Layers: Integrating with fragmented hospital information systems remains a major obstacle. Furthermore, the company maintains strict clinical safety standards through internal clinical testing and works to secure stakeholder buy-in by positioning the AI as a tool to assist, rather than replace, existing staff.

Q&A Insights

• Data Privacy (PII): To protect patient confidentiality, Outcomes AI utilizes HIPAA-compliant commercial agreements and is actively exploring self-hosted LLM solutions to keep data entirely in-house.
• Handling Pauses: To prevent the AI from interrupting patients who pause to think, the system pairs its transcription model with a dedicated end-of-utterance detection model to determine when a speaker has truly finished.
• Adoption and Safety: Voice AI adoption in US healthcare is in its infancy. The technology has only recently matured enough to handle multi-minute conversations reliably. To mitigate liability, parallel guardrail systems prevent the AI from offering direct medical diagnoses, keeping a human nurse in the loop for all clinical decisions.