Click here to view the full RFP

Population Services International
Technology Vendor Artificial Intelligence Coach Development
Request for Proposals

To Provide: AI Virtual Coach Development

Issue Date: August 22nd, 2024

Last Date for Receipt of Clarification Questions by PSI: August 29th, 2024 11:59 PM PST

Last Date for Responses by PSI: September 3rd, 2024 11:59 PM PST

RFP Closing Date and Place: September 12th, 2024 11:59 PM PST

Please direct all proposals/inquiries regarding to this RFP to:
[email protected]

Introduction

PSI is a non-profit organization working to improve the health of people in developing countries. PSI has an annual budget of $400+ million, offices in 60+ countries, and 8,000+ employees. For further information on PSI’s size, scope, structure, and strategy refer to www.psi.org.

Scope of Work

Priority Health Area/Audience
PSI seeks to develop a digital health intervention to support men living with HIV in South Africa. The Virtual Coach is designed to provide psychosocial support, deliver accurate health information, facilitate service referrals, and offer practical assistance to men living with HIV in South Africa. The AI-driven tool will seek to emulate the role of an existing human coach model called Coach Mpilo (www.coachmpilo.co.za).

Goals
To increase access to psychosocial support, health information, and referral to additional services and support for men living with HIV in South Africa.

Objectives (Suggest listing specific activities or a class of activities that are mutually agreed upon and any monitoring requirements.)
Consultant/Vendor will:

1. Identify and tailor an advanced Large Language Model (LLM) to serve as the core intelligence of the Virtual Coach for AI Model Selection and Customization. Evaluate enterprise-grade and/or open-source LLMs such as GPT-4, Llama3, or Claude. Customize the selected model to ensure it comprehends and responds to HIV and TB-related inquiries with a high degree of clinical accuracy, empathy and cultural sensitivity. Fine-tune the model (if needed) to align with the specific needs and context of the target demographic.
2. Equip the AI model with the ability to access and utilize a knowledge base with the latest health information during interactions. Implement a Retrieval Augmented Generation (RAG) system to enhance the LLM’s ability to draw on a curated knowledge base of vetted HIV and TB information. Develop and maintain this knowledge base, incorporating semantic search capabilities for accurate and contextually relevant information retrieval.
3. Develop a robust framework for guiding AI interactions across a range of scenarios through Prompt Engineering and Conversational Design . Create and optimize a series of prompts to steer AI responses, ensuring the Virtual Coach can manage diverse conversation scenarios such as providing emotional support, explaining complex medical concepts, and handling sensitive topics. Incorporate best practices in conversational AI design to enhance user engagement and satisfaction.
4. Enable the Virtual Coach to perform practical tasks, extending its utility beyond conversational support, including but not limited to Function Calling and Task Automation. Integrate function calling capabilities to allow the AI to execute tasks within the conversational flow, such as retrieving and sharing appropriate testimonials, locating and referring users to nearby clinics, setting reminders, or scheduling appointments. Ensure seamless task automation to enhance the AI’s practicality and user experience.
5. Establish a reliable and efficient data handling system to support the AI’s operations, via designing Data Management and Processing Infrastructure. Develop comprehensive data pipelines for processing, structuring, and updating information. Ensure real-time data access and maintain data integrity throughout the system. Implement mechanisms for continuous data ingestion and updating to keep the AI’s knowledge current.
6. Identify potential risks to the AI coach users and implement appropriate guardrails and mitigation strategies. Conduct a comprehensive risk assessment to identify potential harm, including but not limited to misinformation, privacy breaches, and psychological impacts. Develop and implement robust safeguards such as content filtering, trigger warnings for sensitive topics, and clear escalation pathway to human support when necessary.
7. Protect user data and ensure compliance with health sector regulations to ensure Privacy, Security, and Compliance requirements are met. Implement stringent data privacy and security protocols, adhering to healthcare standards such as South Africa’s Protection of Personal Information Act, HIPAA or local equivalents. Develop and deploy anonymization techniques for data used in training, evaluation, or storage. Ensure that all systems are compliant with relevant legal and ethical guidelines, including the World Health Organizations ethics and governance of artificial intelligence for health. Guidance on large multi-modal models.
8. Continuously assess and enhance the performance of the Virtual Coach through the design of an AI Model Evaluation and Continuous Improvement activities. Design a comprehensive evaluation framework to monitor AI performance across various metrics, including clinical accuracy, empathy, cultural appropriateness, and user satisfaction. Implement feedback loops to facilitate ongoing improvements, informed by real-world usage data and evolving project requirements.
9. Ensure the Virtual Coach can effectively communicate in both English and Zulu to provide Multilingual Support for users. Assess and enhance the multilingual capabilities of the selected LLM, with a specific focus on Zulu language support. Implement language model or translation system as needed to ensure accurate and culturally sensitive communication in both languages.
10. Integrate the AI with user-friendly platforms such as WhatsApp and Interactive Voice Response (IVR) systems. Provide backend support for seamless integration with popular messaging platforms and voice-based systems. Ensure the Virtual Coach can operate smoothly across these platforms, maintaining a consistent user experience regardless of the medium.
11. Develop detailed documentation and provide training to facilitate the effective use and management of the Virtual Coach. Produce thorough technical documentation covering all aspects of the system, from model architecture to data handling processes. Conduct knowledge transfer sessions with the project team to ensure they are equipped to maintain and scale the Virtual Coach as needed.
12. Provide continuous technical support and system enhancements post-deployment. Establish a support structure for troubleshooting and system optimization. Offer regular updates and refinements based on user feedback, emerging technologies, and evolving project needs.

Expected Deliverables/Measures
Consultant/Vendor will:

1. Develop a comprehensive workplan.
2. Develop documentation that captures the design and technical specifications of the Retrieval Augmented Generation (RAG) system.
3. Share approved prompts and conversational frameworks in both English and Zulu.
4. Provide documentation of the functional task execution workflows.
5. Share an approved data management and processing infrastructure plan.
6. Provide documentation on comprehensive data privacy and security requirements are met.
7. Develop a project monitoring and evaluation plan including and improvement protocols.
8. Successfully integrate user interfaces with WhatsApp and IVR.
9. Develop documentation to support knowledge management and compile training materials.
10. Develop a fully operational AI Virtual Coach system that meets the requirements above and is satisfactory.
11. Deliver a comprehensive risk assessment and mitigation strategy document, including identified potential risks to AI coach users, implemented guardrails and safeguards, user well-being features, and ongoing risk monitoring protocol.

Language
All documents submitted in response to this RFP, as well as all correspondence in connection
with the RFP, shall be in the English language.

Clarifications
Inquiries concerning this RFP must be submitted in writing, to be received by the date shown on the Cover Page. Such inquiries shall be sent by email, directed to

Email: [email protected]
Subject: Information Request: PSI_DHM20240402

PSI is under no obligation to consider or respond to questions that are not received by that date.

Bid Preparation
Bidders should submit their responses to this RFP to include:
A. Technical Proposal
B. Financial Proposal

A. Technical Proposal
Your proposal should provide basic information about your Company and relevant service
offerings. It should have a section addressing each of the following areas:

(a) Company Profile
(b) Approach

(c) Responsiveness to the Scope of Work
(d) References

Responses should include the following information:
a. Company Profile

• Describe the Company on an overall basis, both nationally and internationally (if
  appropriate)
• Summarize the Company’s qualifications and experience of the activities described in the scope of work
• Provide a detailed track record of previous similar work undertaken

b. Approach

• Provide a description of the Company
• Describe the Company’s experience and capacity to design and implement digital health intervention that utilize artificial intelligence.
• Describe the Company’s capacity to provide English and Zulu language support.
• Describe the Company’s approach to product development.
• Describe the Company’s approach to project management.
• Describe the Company’s approach to quality improvement.
• Describe the key personnel to be utilized on the engagement along with CVs of key personnel, indicating their degree of expertise and prior experience as relevant to this engagement.
• Provide any other information to demonstrate the Company’s capability in relation to the Scope of Work.
• Identify the point of contact who will be available for consultation or discussion.
• Describe any current or past relationships your organization may have with PSI, and if it is a potential conflict of interest. If there is a potential conflict of interest, then please explain how this risk will be mitigated.

c. References:

• Submit three client references, including contact information.

B Financial Proposal
Bidders pricing should be based on the following:

For Services performed during the Term, Consultant will be paid a fixed fee, to be determined during the contracting process (the “Fixed Fee”). Applicants should provide a budget and justification. The Digital Health and Monitoring (DHM) department issues deliverable-based contracts. Deliverables and payment amounts will be finalized during the contracting process.

As part of the RFP process, PSI may be providing confidential information to bidders. Consequently, bidders must complete the attached Confidential/Non-disclosure Agreement (Part 4) to participate in this process. Signed Confidential/Non-Disclosure Agreements must be submitted with the proposal.

Bid Submission
All proposals must be in writing, in the English language, and manually signed and dated by an authorized employee of the Bidder. They may be emailed as shown on the RFQ cover page.

Proposals may not be altered, corrected or withdrawn after the Date of Receipt, except that PSI, at its sole discretion, may permit correction of arithmetic errors, transposition errors, or other clerical or minor mistakes, in cases in which PSI deems that both the mistake and the intended proposal can be established conclusively on the face of the proposal. Other than the mistakes listed in the previous sentence, no mistakes alleged by a Bidder after the Date of Receipt will be permitted to be corrected.

Proposals must be valid for at least ninety (90) days from the Date of Receipt. #li-remote #li-lm1