Multilingual Conversational AI
LLM-Based Healthcare Chatbot for Health Services
Led the product direction for a multilingual, LLM-powered healthcare assistant designed to reduce repetitive public inquiries, improve access to trusted health information and automate common interactions such as vaccination guidance and appointment support.
Industry
Healthcare / Public Sector
My Role
AI Product Manager
Team Size
Cross-functional team across Product, AI/Data Science, Engineering, UX and Policy
Duration
Sep 2024 – Mar 2025
1. Background / Context
A public health organization was handling high volumes of repetitive inquiries related to vaccinations, health guidelines and appointment scheduling. Citizens relied heavily on phone support, placing pressure on call center teams and making it difficult to provide timely information at scale. The opportunity was to use conversational AI to provide 24/7 access to trusted health information while allowing human agents to focus on more complex cases.
At a Glance
Solution Type
Multilingual Conversational AI
Platform / Environment
Healthcare Digital Services
Technologies
Falcon LLM, Jais LLM, LangChain, Azure AI, REST APIs
User Impacted
Patients, Citizens, Call Center Teams & Public Health Officials
2. Challenge
• High volumes of repetitive health and vaccination inquiries were handled manually.
• Call center workload contributed to longer response times and delays.
• Citizens needed access to reliable information outside normal support hours.
• The solution needed to support multiple languages for a diverse user population.
• Healthcare information required strong accuracy, privacy and compliance controls.
• The experience needed to integrate with existing systems for appointments and real-time information.
3. Solution Design
• Conversational access to vaccination information and health guidelines.
• Appointment booking and reminder support.
• Multilingual interaction including Arabic and English.
• Integration with healthcare systems through APIs for current information.
• Content and policy validation for healthcare responses.
• Post-launch monitoring of missed queries and chatbot performance.
Key Finding
The highest-value use case was not replacing human support, but automating repetitive information requests so human agents could focus on complex citizen needs.
Identify High-Volume Queries
Define Priority Intents
Design & Validate Conversations
Integrate & Launch MVP
Monitor & Improve
4. Product Decision
Prioritized high-volume vaccination and appointment use cases rather than attempting to automate every healthcare inquiry, enabling the MVP to be delivered within six weeks of discovery.
Focus the MVP
Human Support for Complexity
Positioned conversational AI around repetitive and predictable interactions, allowing human support teams to remain focused on more complex citizen needs.
Multilingual by Design
Included multilingual capability as a core product requirement to make the service accessible to a broader and more diverse population.
Accuracy Before Scale
Worked with content and policy experts to ensure responses were accurate, localized and aligned with healthcare requirements before expanding the experience.
5. Outcomes & Impact
35%
Reduction in support tickets within the first month
10,000+
Queries handled during early rollout
6 Weeks
From discovery to MVP delivery
Multilingual
AI-powered healthcare access across key languages
6. Key Learnings
What Worked Well
• Using support-volume data to prioritize the highest-value use cases.
• Keeping the initial MVP scope tightly defined.
• Bringing policy, content and technical teams into product discovery early.
• Validating prototypes before wider rollout.
• Monitoring real conversations after launch to identify missed queries and improve performance.
What We Would Do Differently
• Define a more structured LLM evaluation framework earlier in development.
• Establish detailed fallback and escalation journeys earlier for questions the assistant cannot confidently answer.
• Build stronger self-service analytics for content, intent and language performance from the start.
• Introduce more systematic user feedback collection alongside operational metrics.
Key Takeaway
In high-trust environments like healthcare, conversational AI succeeds when speed and automation are balanced with accuracy, accessibility, human oversight and clearly defined product boundaries.