The Policy Innovation Lab has developed Africa’s first AI for policy platform, now accessible to registered users via https://policyinnovationlab.tech/login. Designed specifically for policymakers in South Africa’s government, the platform offers secure, private access to a suite of AI tools built on Large Language Models (LLMs).

It provides a single point of entry for a range of fit-for-purpose tools, with privacy and security embedded throughout. The platform stores only the minimal data required to maintain security and ensure compliance, helping to build trust in the use of LLMs in government, where use cases are often politically or operationally sensitive.

The tools use a design in which the LLM is provided with relevant source material as a basis for its outputs. This content is either uploaded by users or gathered automatically through data collection methods. The result is a significant reduction in hallucination and bias, with context-specific prompts that improve accuracy and relevance.

Effective use of the tools depends on informed human oversight. Understanding how the AI works and critically assessing its outputs remain essential to keeping problematic content out of decision-making. To support this, the Lab offers training and capacity-building for government officials using the platform. The tools are as follows:

Media brief generator

The media brief generator allows users to automatically obtain an overview of recent social media posts and mainstream media articles, identifying trends, priority issues and public sentiment related to a topic of their choice and over a period of up to 7 days. LLMs provide structured analyses of media data and how different groups or actors respond to government actions. A use case of this tool is that some action by the government generated media activity and they want a quick and yet broad update on how people are responding in both local and global mainstream media and social media. This addresses a need that traditional media analysis, which provides insight into medium- and long-term trends, typically does not.

Policy brief generator

This tool creates a draft policy brief on a topic provided by the user by analysing academic research. The tool retrieves open-access academic articles published over a specified date range related to the user’s search query. A Large Language Model (LLM) then summarises each of these articles, extracting information that may be relevant to policymakers while removing unnecessary information such as methodologies and bibliographies. These summaries are then used to create a policy brief following a standard structure. The LLM will also make a statement regarding how relevant the retrieved research is to the search query. This provides policymakers a starting point for generating policy briefs or policies, and getting up to speed on academic research in policy-friendly language.

Policy summarisation and comparison

These two tools utilise LLM technology to summarise documents or compare them. The LLM has default instructions that help it to structure outputs and keep them relevant; however, users may also provide their own instructions. A use case for these tools is policymakers trying to identify which other policy or strategic documents align or clash with the policy they are working on. They can do this by prompting the summarisation tool to identify what other policies say about their topic, and prompting the comparison tool to identify congruencies or incongruencies between their draft policy and national strategies.