A team of six AI agents that answers questions about Dubai law, using RAG and GraphRAG on AWS Neptune. Every answer points to the articles behind it. Built during my AI engineer internship at E-tafakna.
Yes, if both parties agree or if the contract allows it. Check the penalty clauses before you act.
Sends the question to the right specialist agent.
Finds related laws and articles, such as 14 for one query.
Drafts the answer from the laws it found.
Checks every citation before the answer goes out.
The core challenge of E-tafakna was building a system that lawyers could trust. Standard RAG (Vector Search) wasn't enough because laws reference other laws. We implemented GraphRAG using AWS Neptune to map relationships between articles. When a query comes in, our Router Agent assigns it to a specialist (e.g., Labor Law Agent), which traverses the graph to find the exact legal pathways before drafting a response.
As an AI Engineer Intern, I owned the multi-agent pipeline. I designed the graph schema, wrote the Cypher queries to extract context, and built the Reviewer Agent that strictly verifies every citation in the generated response against the original legal text to prevent hallucinations.
Ask a question about the tech stack or my role in the project.