A national research agent for The Youth Company

The situation

Youth.qa is a platform run by The Youth Company for young people in Qatar aged 13 to 29: jobs, internships, programmes, events and scholarships in one place, in Arabic and English. A platform like that lives or dies on two things: fresh content every day, and institutions willing to list their opportunities on it. Both are work, and the team is small.

This is a partnership. Cliqom works alongside The Youth Company rather than for it.

The constraint

Volume without a newsroom. The platform needed articles published on a daily rhythm and hundreds of institutions contacted and followed up, in two languages, by a team that also had a platform to run. Anything we built had to produce publishable Arabic, not translated Arabic, and nothing could go out without a person deciding it should.

What we built

Two automation engines.

The article engine takes sources the team has approved, drafts a piece in Arabic and English with an AI model, checks it for facts the team has flagged as sensitive, and places it in a review queue. A person reads it, edits it, and publishes it. Nothing publishes on its own.

The outreach engine takes a list of institutions, writes a personalised first message, sends it, waits, follows up on a schedule, and records every reply in the team's CRM. Replies that need a human answer are routed to one.

How we handled the hard part

Automating the drafting is easy. Automating the judgement is a mistake. The line we drew: the machine may write, sort, schedule and remind; a person decides what is published and what is sent. Every draft carries the sources it was built from so the reviewer can check rather than trust. Arabic is generated as Arabic, then read by an Arabic speaker before it is seen by a reader.

What changed

The team reviews rather than writes, and follows up on replies rather than chasing silence.

What we would tell the next client

Decide first what the machine is never allowed to do, and build that in as a rule, not a setting. Then automate everything on the other side of the line as hard as you like. The rule is what lets the team trust the output enough to actually use it.