Case study · Government · 2026
Kelmet Masrكلمة مصر
Every citizen proposal from Egypt Forum 2026, read in full
Vega Data built Kelmet Masr in collaboration with Egypt's Ministry of Foreign Affairs. It turns 5,620 published proposals, about 1.8 million words written by 3,535 citizens, into an analysis that the ministry, the Prime Minister's office and any citizen can search and read.

- published proposals read in full, with no sampling
- 5,620
- published proposals read in full, with no sampling
- proposals ready to implement
- 67
- proposals ready to implement
- ideas raised independently by three or more citizens
- 200
- ideas raised independently by three or more citizens
- government commitments set beside citizen proposals within a day
- 72
- government commitments set beside citizen proposals within a day
The challenge
Egypt Forum 2026 ran for six days from 19 October 2026, with 26 sessions. At the opening, the Prime Minister said the forum had received more than 16,000 ideas and proposals from citizens. Of these, 5,620 were published on the forum's platform. They were written by 3,535 people and run to about 1.8 million words.
That is more text than a team can read closely in the days a forum lasts. Reading a sample would miss the ideas that many citizens raise independently of one another.
The collaboration
Vega Data developed Kelmet Masr in collaboration with Egypt's Ministry of Foreign Affairs, under a written agreement. The work began before the forum opened. The ministry uses the platform in its work, and the Prime Minister's office has used it to review what citizens proposed.
The same analysis is open to everyone at kelmetmasr.com, in Arabic and English. Vega Data builds, owns and operates the platform. Every proposal links to its official page on egyptforum.gov.eg, which remains the reference.
What the platform does
- Reads everything. All 5,620 published proposals are read in full, with no sampling.
- Sorts by readiness. Each text is checked for ten elements of a workable proposal, among them the problem, how it would be implemented, cost and funding, indicators, a timeline and risks. Support from other citizens is measured as well. Together they place every proposal in one of four groups, and 67 are ready to implement.
- Finds shared ideas. 200 ideas were each raised by three or more citizens writing separately. Together they account for 1,029 proposals.
- Organises the detail. Proposals can be browsed by 19 sectors, 27 governorates, 7 national programs, the 32 ministries and public bodies they address, and 13 groups of citizens.
- Sets proposals beside the government's response. The Prime Minister announced the forum's recommendations at the closing session on 24 October 2026. Kelmet Masr reads them as 72 commitments in 20 themes, 22 with a stated deadline, and shows each theme next to what citizens proposed.
- Answers questions. Thoth, the platform's assistant, answers in Arabic or English from the proposals' own texts and links every proposal it cites.
Ready to implement67
Supported, needs study846
Complete, needs visibility251
Early ideas4,456
How it runs on Microsoft Azure
Kelmet Masr is built and hosted on Microsoft Azure. Thoth is built on Azure AI services. Vega Data's software does the reading, scoring and grouping, and every figure on the site is computed automatically from the proposal texts.
Results
- All 5,620 published proposals read and classified.
- The analysis of the government's 72 commitments online within a day of the closing session.
- Used by the Ministry of Foreign Affairs and by the Prime Minister's office, and open to every citizen.
- Published under CC BY 4.0, so anyone can reuse the analysis with credit.
Platform and infrastructure
- Azure AI Foundry
- Hosts the Azure OpenAI models behind Thoth, the assistant that answers in Arabic or English from the proposals' own texts.
- Azure OpenAI embeddings
- text-embedding-3-large finds the proposals that answer each question, so every answer can cite them.
- Azure Virtual Machines
- Run the platform and the reading, scoring and grouping pipeline (France Central).
- Managed disk snapshots
- Point-in-time copies of the production data.
