Artificial Intelligence (AI) agents are transforming healthcare operations by enhancing productivity and expanding capacity.  These intelligent assistants support administrative tasks, facilitate patient engagement by enabling seamless multi-model communications such as voice and text.  For this Call for Innovation, it will primarily focus on patient and community management and healthcare operations optimisation for a start. The agentic AI tasks could be varied, but they mainly assist healthcare professionals in engaging and educating patients or residents by conducting appointment (procedure) reminders, medication counselling, pre- and post-procedure follow-up care, and educating/enrolling residents in specific programs or activities.  

Key considerations involve the extent of parameters and guardrails to ensure optimal data privacy and security, ethical implications for addressing clinical risks associated with AI-generated recommendations and responses, and seamless integration into existing workflows and backend source systems, such as Next-Generation Electronic Medical Records (NGEMR), for patient safety, minimise disruption, improve compliance, and enhance overall continuity of care. 

How might we leverage the current agentic AI technology to support Singapore healthcare professionals in effectively managing routine voice-driven outreach tasks in a timely, safe, autonomous and engaging manner?

The proposed Agentic AI technology solution for this Proof-of-Concept (POC) trial should cover the following 3 types of use cases (refer to Table 1 below): 

Type of Use Cases	Type A	Type B	Type C
Use Cases	Medication counselling	Pre- and post procedure counselling	Population health outreach and community nursing
Technology	Agentic AI (voice-based/voice-driven automation)
Specific Requirements	i) Educate and guide patients (e.g., medication adherence and understanding)  ii) Provide follow-up care	i) Address patient concerns (e.g., appointment management, urgent enquiries)  ii) Offer directions to patients (e.g., procedure-specific instructions)  iii) Provide follow-up care (e.g., reassurance to caregivers)	i) Appropriately nudge (offer reminders) based on profile and enrolled programs    ii) Adapts messaging based on demographics, language preference, and cultural considerations  iii) Educate and guide local resident
Capability	i) Convert audio conversation into text transcriptions (call logs)  ii) Call escalation and routing abilities  iii) Low-code/no-code self-help interface for AI agent management (i.e., create new AI agents, edit/manage/configure workflows)  iv) Feedback opportunity – to identify patterns and address emerging concerns (i.e., health-related, operational-related e.g., satisfaction levels)
Types of Users	Pharmacists	Nurses	Nurses/Allied Health
Target Audience	Patients   (i.e., inpatient discharge and   home delivery)	Patients undergoing/ completed procedures   [e.g., Ear, Nose & Throat (ENT), Ophthalmology, Endoscopy, Obstetrics and Gynaecology (O&G)]	Adult (aged 18 and above) Singapore (local) residents who can benefit from preventive health interventions   (e.g., at risk of chronic diseases - hypertension, diabetes, cardiovascular conditions or eligible for national screening programs like cancer screening)
Language Support Requirements	Multilingual   i) Mandatory: English and Mandarin  ii) Optional: Tamil and/or Malay

Table 1: Use Case Types- For detail of the use cases, please refer to Annex A: Use Cases Summary here.

1. Join us on this discovery and collaborative journey through HealthX to explore and build newer, valuable, technology-enabled solutions to:

i.Boost productivity:Automating routine and repetitive tasks (e.g., medication, pre- and post -procedure counselling, and educating Singapore residents about preventive health programs and measures) can lighten healthcare professionals’ burden and allow them to handle more complex cases or other essential duties. 

ii. Improve the quality of patient care: Standardising patient-provider conversations, i.e., medication counselling, pre- and post-procedure counselling, nudging, and educating, enables patients, caregivers, and local residents to receive information at a time convenient to them (e.g., at the comfort of their homes) preferable in a local context that are familiar to them. 

2. The proposed solution for this POC shall fulfil the following capabilities to be considered as a suitable candidate: 

i. Multi-language support: Ability to support multi-languages (both English and Mandarin at the minimum). Added advantage if the solution could support other mixed languages (e.g. Malay, Tamil, etc.) for Singapore context within the same conversation (task) log. 

ii. User friendliness: The proposed solution should be intuitive for users to use (low-complexity and user-friendly) with minimal guidance and support (without elaborate setup or configurations). 

iii. Interactive conversational AI: The proposed agentic AI solution should possess multimodal literacy capabilities, emotional intelligence, proactive engagement abilities, personalisation and contextual awareness via Natural Language Understanding (NLU) and Natural Language Processing (NLP), reasoning and contextual reply, and can handle interruption to fulfil following scenarios: 

a. A voice assistant for real-time counselling with the ability to answer patients’ questions on medication use and identification, possible risks and procedure-related complications and precautions.  

b. Any reminders or follow-up calls that can boost compliance will be an added advantage. Incorporation of the solution with any existing patient fronting platform, interactive videos, or visual aids will also be an added advantage. 

c. A voice assistant that can accurately and empathetically answer patients’/residents’ questions while providing the relevant guidance and encouragement to complete the assigned actions (e.g., enrol in a program, share about their condition, assist with vaccination appointment scheduling). Any reminders or follow-up calls that boost compliance will be an added advantage. Incorporating the solution with any existing patient fronting platform or interactive videos or visual aids will also be an added advantage.  

iv. Personalised guidance: The proposed solution should be able to provide specific information, for example: 

a. Automated pre-procedure check in, reminders for fasting instructions, medication adjustments, procedure specific instructions and arrival times. 

b. Automated post-procedure check in via voice or chat to assess pain, complication or adherence to post-procedure treatment plan. Counsel on common signs and symptoms of post procedure related complications and when to seek medical attention.

v. Escalation protocols: The proposed solution must be able to transfer queries to a clinician when not able to unable to answer or if a high-risk scenario is identified. The AI solution should be able to summarise the unresolved queries to the human operator before handing over the call. 

vi. Platform-level enabler: A self-service, low-code/no-code interface to enable users to easily create new AI agents to meet their specific requirements (i.e., use cases/workflows) without requiring extensive technical expertise. 

3. The proposed solution for this Proof-of-Concept (POC) shall fulfil the following technical requirements: 

i. Data residency: For solutions that leverage their proprietary Large Language Model (LLM) hosted at their native location for the processing of audio to text and summarising capabilities, the solution provider may have to demonstrate that the audio and transcript text file do not reside at their native location. 

ii. Data security and infrastructure requirements: The solution must adhere to privacy standards (i.e., PDPA) to safeguard patients’ information and the Security and Infrastructure Requirements on Controls for Trial & Pilot Environment (Refer to Annex B for more details). 

iii. AI-Safety: Added advantage if there is AI-safety services/features to detect and block sensitive health information and personal identifiable information. 

iv. Natural Language Processing (NLP): The solution must be able to handle Singapore’s diverse languages, dialects, and accents in Singapore. 

v. Integration with other health systems and applications: The solution needs to be able to integrate with other healthcare systems and applications, especially to receive the latest medication list. It should be able to tailor counselling based on different procedure preparation, patient history, allergies, and concurrent medications.  Once of such core backend system is with the Next-Generation Electronic Medical Records (Epic).  Solution providers need to have either proven track records in Epic integration or express explicit commitment to establish Epic integration by the time of actual implementation of the solution after the trial. For the latter, it is the sole responsibility of the solution provider to work directly with Epic to get their integration endorsed by Epic. 

vi. Leverage on Synapxe Tandem platform infrastructure: Tandem is a Generative AI platform developed by Synapxe specifically for public healthcare professionals. The proposed solution can explore leveraging on Tandem platform infrastructure for basic security hygiene such as active directory authentication for public healthcare users, AI safety features, AI models hosting or secure access to cloud service provider (CSP) LLM. Refer to Annex C on Tandem platform a high-level architectural design. 

4. Overall collaboration requirements: 

i. Scalable: The proposed solutions should consider a broader plan to scale across other healthcare institutions and settings and consider the need to integrate and interface with the various relevant systems. 

ii. Cost-effectiveness: The proposed solutions must be cost-effective and beneficial to the public healthcare in Singapore. To support time motion study (if needed) to justify/support the business case and return on investment (ROI). 

iii. Implementation Timeline: This POC collaboration is scoped to address the challenge statement, and the trial completion period is to be capped at not more than 3 months (including the setup of the trial).  

iv. Cost: The proposed solution should include indicative cost (business model) for the short-term implementation and long-term scaling prospects.  

 

For full details on the use cases, please click here.