An open-source AI platform from Sharif University's Advanced ICT Research Institute. On one side, an AI assistant. On the other, the compute, models and APIs that organisations build on. One design system covers both.
Two kinds of user on one platform: people who want an answer from an AI assistant, and teams who build their own AI services on its compute, models and APIs.
I led the platform's design part-time, for nine months. I built one design system across every admin panel and set the product's direction until its teams could carry it on with the system alone. Once the heavy design work was done, I moved to an advisory role, and I still own design there.
Every screen here is built from the same parts: one sidebar with search at the top and the account switcher at the foot, one card, one pair of filled and outlined buttons, and the help button in the same corner. A person asking the assistant a question and an engineer creating a cluster are in the same product.
Teams that build on the platform find everything in one section: keys, a place to test calls, SDKs and the docs. The public API reference keeps its Persian prose right to left but runs code, keys and parameters left to right. The request schema sits in the middle and copyable samples beside it.
Accounts belong to organisations, so sign-up ends on one question: which one are you with? By then a person may already be a member somewhere, hold an invitation, have asked to join one, or have created one that an admin hasn't approved yet. The picker shows all of it in one list. Adding an organisation means creating one or asking to join an existing one, and either way it waits for an admin.