Presidential Library AI Research Assistant

A historical archive of this depth is only as useful as it is discoverable. Researchers and visitors exploring Syria's democratic era needed a way to ask specific, nuanced questions across thousands of documents without knowing exactly where to look. The challenge was building an AI layer that could handle that complexity faithfully, with the right guardrails to ensure responses stay grounded in the archive's actual content.

We built a RAG-based assistant that indexes a curated collection of the library's documents, making the archive queryable through a natural language interface embedded directly into the website. The response harness was designed to keep answers accurate and anchored to source material, so the assistant serves as a reliable research tool rather than a generative one.

Built on AWS Bedrock with Claude, a curated selection of the library's documents and records are extracted, chunked, and stored in Amazon S3, feeding an Amazon Bedrock Knowledge Base. Visitor queries are handled through the AWS Bedrock RetrieveAndGenerate API and the assistant is embedded into the website as a JavaScript component.

The library's archive is now as easy to query as it is comprehensive. Researchers can surface specific documents, dates, and decisions through conversation, making one of the most significant records of Syria's parliamentary history more accessible than ever before.