Skip to content
Hitoriah

Selected work

Systems built for real businesses, in Morocco and beyond: manual work removed, sales pipelines made legible, and ad creative produced at the volume testing actually demands.

Every project runs the same sequence: find the bottleneck, prove the fix with a working prototype, launch, then improve on evidence. Organisations are not named, and each entry describes what was built rather than performance figures.

  1. 01

    E-commerce ad engine

    The number of angles worth testing was capped by how many variations could be produced, and every placement and every market wanted its own cut. We built a repeatable way to turn approved source images into a set of video concepts, adapted to each platform format and to the language of each market. Nothing reaches an ad account without a person approving it.

    AI video advertising
  2. 02

    Sales CRM workspace

    Enquiries arrived through WhatsApp, email and forms with no shared record, so some were answered twice and others not at all, and nobody could say what stage a deal was at. We built one intake and an explicit pipeline where every deal has a status, an owner and a next step. Stages advance from what actually happens instead of waiting for someone to remember, follow-ups fire on their own, and reporting shows what is open and what has stalled.

    Custom CRM
  3. 03

    Operations automation

    The same information was typed by hand into several systems, with the time and the mistakes that implies, while someone else spent their day chasing status and rebuilding a recurring report from exports. We connected the systems so data is entered once, added checks that stop a bad value where it is entered, and let status updates and the report generate themselves. Anything outside the rules goes to a person.

    Custom integrations
  4. 04

    Support AI assistant

    Request volume had outgrown the team: every ticket was read by a human just to route it, and the same answers were rewritten constantly. We built triage that routes automatically and drafts replies grounded in the client documentation and in past resolutions. The assistant answers only from sources the client has approved, holds no access beyond what its scope requires, and declines rather than improvises when a question falls outside it. An agent reviews and sends.

    Process automation

What keeps these safe to run

A person approves anything that leaves the system, exceptions go to a human rather than being guessed, and what the system did can be reconstructed afterwards. Access, documentation and ownership are handed over at launch in every case. What differs is what comes after: some teams take on maintenance and administration themselves, others want the functionality and leave the running to us.

Have a similar use case?

Let's scope a solution around your tools, constraints and objectives.