Introduction
As the Generative AI Strategist at the World Health Organization, I have been leading the design and deployment of large language model (LLM) architectures to fundamentally transform how WHO processes knowledge, automates reporting, and scales global insight generation. My primary mission is to transition WHO into an AI-first organization — one where decision-making, policy communication, and reporting are streamlined by AI, enabling rapid, reliable, and strategic responses in global health governance.
Our focus has been on leveraging LLMs to optimize WHO’s most information-intensive processes, including donor reporting, scientific briefings, assembly speech transcription, and Director-General office intelligence workflows. By embedding AI across departments, we have established new ways of producing, structuring, and delivering health intelligence at scale — helping WHO fulfill its mission as the global authority on health.
Target Group
Primary: Senior WHO Executives, Chief Scientific Office, Director-General’s Office, and Strategic Planning Units
Secondary: Data Analysts, Report Writers, Technical Officers, Policy Teams, Donor Relations, Communications Teams
Project Objectives
AI-Native Strategic Intelligence
Transform WHO into an AI-native organization by embedding LLMs into core operational and strategic workflows, enabling real-time policy evaluation, faster response cycles, and informed decision-making.
Knowledge Workflow Automation
Automate the generation, structuring, and summarization of health reports, transcriptions, and analytical products, drastically reducing the cognitive load on WHO staff and improving global reporting turnaround.
LLM-Driven Reporting Infrastructure
Develop internal AI tools — including Microsoft Word Add-ons and internal GPTs — to support report authors, editors, and country representatives in their drafting, reviewing, and publishing workflows.
What We Did for WHO
Comprehensive AI Architecture Deployment
Using the 8-layer AI-first architecture developed with Metamatics, we implemented a fully integrated LLM-based system covering:
- Document Ingestion & OCR: Automated processing of scanned PDFs and multilingual content from assemblies and regional offices
- Speech Transcription & Summarization: Real-time processing of country-level assembly speeches, enabling searchable records, diplomatic tone detection, and thematic extraction
- Donor Report Automation: Auto-synthesis of donor-specific outputs aligned with programmatic indicators and reporting deadlines
- Strategic Dashboards: For executive offices and policymakers, highlighting risks, program impact, and recommended actions
AI-Enhanced Reporting & Knowledge Generation
- Word Add-ons: We developed AI-powered Word extensions that assist WHO staff in summarizing, editing, and standardizing policy and scientific documents.
- Assembly Intelligence Layer: Transcription and annotation tools for high-level global speeches, designed for the Executive Board and World Health Assembly.
- Prompting Systems: Designed prompting workflows for Director-General briefings to summarize cross-departmental input and structure it into action points.
Governance and Compliance Integration
- Aligned all tools with global health ethics, including bias checks, data privacy measures, and compliance with AI governance standards (e.g., WHO AI guidance, GDPR, ISO 42001).
- Integrated observability and explainability layers to ensure that every insight or recommendation produced by AI could be traced, audited, and interpreted by human experts.
Key Challenges
Volume and Diversity of Inputs
WHO handles a wide variety of report types, languages, and document structures across regions. Our models had to adapt to both standardized formats and ad-hoc, mission-specific documentation.
Real-Time Use with High Stakes
The AI tools had to perform reliably in high-pressure environments such as international assemblies or DG briefings. This required not only robustness, but seamless integration into workflows already used by high-level staff.
Balancing Precision and Flexibility
We needed to support both structured data-driven reporting (e.g., donor KPIs) and more discursive narrative summaries, requiring hybrid models combining NLP precision with creative language generation.
Outcomes and Impact
- 50% reduction in time required for multi-author report generation
- First-ever real-time assembly speech annotation layer deployed
- Adoption across 4+ core departments, including Chief Scientist Office, Strategic Planning, Donor Relations
- Enhanced consistency and clarity in cross-regional and multilingual reporting
- Positioned WHO as a global AI governance example, with compliance-by-design workflows embedded into all systems
Conclusion
My work at WHO reflects a broader transformation: not simply adopting AI, but architecting a future where AI is foundational to how global institutions generate, evaluate, and act on knowledge. We are laying the groundwork for a self-learning organization, where information flows intelligently and decisions are guided by insight at scale. With this AI-first shift, WHO is better equipped to lead the world in health — not just through expertise, but through intelligence infrastructure that matches the complexity and urgency of its mission.
