A citizen AI assistant for Egypt's national e-services portal

In 2024 MCIT was mandated to modernise the Digital Egypt portal (digital.gov.eg) with machine learning and generative AI. The portal serves over 30 million active users with more than 150 services from 12 ministries, from official documents and civil status to traffic, courts and payments.

Three obstacles stood in the way:

• Fragmented data. Each ministry’s services ran on separate systems, under strict privacy and security rules.
• Services that were hard to find. Citizens describe their needs in colloquial Egyptian Arabic (“عايز أجدد رخصة العربية”), while services carry formal names, so search and menus often failed.
• No room for invented answers. A government assistant cannot point to a service or a link that does not exist.

My first step was to agree measurable phase-1 targets with stakeholders: shorter citizen request processing times and lower civil service costs.

I led a team of data scientists and engineers across three workstreams:

• Data foundation. I set a data strategy tied to the mandate, built on data integrity, security and scalability, and the team brought the ministries’ services into one data warehouse, hosted on premises under the portal’s privacy and security rules.
• Citizen AI assistant. We built an assistant that understands questions in Arabic or English, answers in Egyptian Arabic, and links users straight to the right service. A three-step pipeline translates the question, retrieves the matching service from a curated catalogue, and writes an answer based only on that content. When nothing matches closely enough, it gives a standard reply instead of guessing. Conversation memory handles follow-ups such as “what documents do I need?”, and every answer returns service codes the portal can act on.
• Recommendation engine. The engine suggests related services. For example, a user asking about a service that needs a valid passport is directed to passport renewal.

• Citizen request processing time cut by 40%, with civil service costs projected to fall by 30%.
• On the web portal, one question in colloquial Arabic returns an accurate answer with direct links, across more than 150 services.
• Answers are limited to the official catalogue; a match threshold and a standard fallback reply prevent invented services.