| Issue/Objective |
Electronic evidence bases have made available various sources of evidence to facilitate evidence-informed decision-making processes. Generative Artificial Intelligence's intelligent retrieval mechanisms could ease access to massive evidence in the electronic evidence base. We used a pragmatic design approach to design & develop Dr-RES, a fit-for-purpose digital evidence base that enhances policymakers' experiences with a digital evidence base using next-generation technologies. |
| Methodology/Approach |
The first phase included developing the infrastructure, which was informed by the team's 15 years of experience implementing the rapid response service for EIDM. The database's multi-layered infrastructure leverages Generative AI based on Open AI's text embedding models and stores the metadata in a pinecone vector database optimised for high-dimensional similarity searches. Parameters included publication titles, abstracts, sources, and outcome classification using the sustainable development goals. Evidence sources were processed using Python libraries to ensure compatibility with the analytical framework. |
| Results |
Dr-RES integrates data from policy-relevant evidence syntheses, including evidence maps and rapid evidence syntheses. The initial repository includes more than 7,000 local studies conducted in Uganda from 2010 to 2023 and rapid evidence syntheses. Outputs include Generative AI quick policy evidence summaries for each report and engagement with the whole evidence base. |
| Discussion/Conclusion |
The Generative AI outputs reduced the time spent retrieving and summarising evidence to less than one minute. Users indicated that the insights from the AI-generated summaries were relevant and enabled faster reviews on different topics. However, the users cautioned that it was important to add a verification mechanism, such as a human in the loop.
The second phase will include engagement with different interests to i) integrate a scalable infrastructure for knowledge users to publish their policy evidence briefs and ii) allow real-time language translation to French and analytics for the scope of available evidence on a policy-relevant |
| Presenters and affiliations |
Peter Mulindwa The Center For Rapid Evidence Synthesis |
