AVA: Making World Bank knowledge searchable with AI

Problem
The World Bank produces thousands of reports and publications each year, yet users often struggle to find relevant insights quickly. Traditional keyword searches return too many results, often with low relevance. Teams needed a faster, more focused way to access reliable, high-quality knowledge rooted in the Bank's own research.

Solution
I contributed to the design and refinement of AVA (AI + Verified Analysis), an AI-powered knowledge tool built to search and synthesize World Bank content. Developed in collaboration with the Nouswise team, AVA provides concise, source-linked answers based on a curated library of official reports and publications.

Key Features
- AI search restricted to World Bank-authored documents
- Natural language query support with multilingual capabilities
- Fully cited responses that link directly to original sources
- Built-in writing assistant for briefs, summaries, and concept notes
- Dashboard to monitor usage trends and common queries

Results
- Enabled faster access to high-quality insights across departments
- Improved user trust by delivering transparent, cited responses
- Supported operational work by reducing the time spent navigating repositories
- Positioned AVA as a model for AI-supported learning in global institutions

My Role
- Identified Nouswise as the most suitable AI solution through a review of market options and tools
- Led internal vetting to align the tool with the World Bank’s institutional needs
- Refined filters to ensure results were limited to official, high-quality documents
- Wrote user guidance outlining what AVA can and cannot do, helping teams apply the tool responsibly
- Compiled feature requirements for the user dashboard, including metrics for search behavior and engagement patterns
- Provided a comparative analysis of similar tools to support strategic positioning and adoption