A generative-AI SaaS designed from scratch in a one-month sprint
GSRAW is a data service specialised in the real-estate sector. In 2024, at the very start of generative AI, I was tasked with designing Copilot, its tool for creating each brand's marketing and communications content in its own voice. We worked as a duo —design and full-stack development— and in a one-month sprint everything shipped, from the UI kit to the landing.
0 → 1
Product, UI kit, flows and landing designed from scratch, with no prior system.
1 month
A single sprint, as a duo with a full-stack developer.
5 tools
Launch, social media, press releases, blog articles and translator.
The brief
GSRAW is a data service specialised in the real-estate sector, expert in developing data and artificial-intelligence applications.
In 2024, at the very start of generative AI, the team wanted to turn that expertise into a product: GSRAW Copilot, an artificial-intelligence tool for marketing and communications in the real-estate sector.
The brief was a full 0→1 in one month — product, UI kit, architecture, navigation and landing — working as a duo: me as lead Product Designer, alongside a full-stack developer.
The challenge
Creating marketing and communications content in the real-estate sector eats up a lot of time and resources, and companies need to produce quality content fast to stay competitive.
The obvious alternative had a catch: using generic tools like ChatGPT makes communications lose their personality against the rest of the market.
Copilot attacked exactly there: generating text with AI, but with each company's brand voice, drawing on its document history and on continuous learning.
The product
The heart of the product is the templates. Each type of content —press releases, blog articles, social posts, translations— calls for a specific structure, voice and style, and the user creates and edits the templates they need to cover any document.
Each template feeds on a description that acts as an instruction, a pre-defined template with the document's fields and output examples taken from the company's own files.
The most interesting piece is the re-training: every correction on a response can be saved into the template, tuning the content to the desired tone and improving the tool with use.
Copilot is also multi-user: administrator, creator and editor roles, member management and control of each team's tokens — the available usage, its limits and the APIs they come from.
From scratch: flows, UI kit and architecture
There was no prior product and no system to lean on. In the Figma file I designed the full journey: login and onboarding, team invitation, content creation and translation, template editing, account settings and error states.
Underneath, a custom UI kit — buttons, inputs, menus, modals and editor components — that holds up every screen in the product.
The landing
The sprint also included the product's landing: value proposition, features, pricing and free trial, with the demo as the centrepiece.
Design note
In 2024, designing on generative AI meant working almost without established reference patterns. The answer was to ground the interface in concepts the user already masters — templates, examples and corrections — so the model can be understood, used and corrected without friction.
Design note
