AI
SabiPT
AI built for individuals and businesses who desire intelligence that connects with their roots.
- Role
- UX Designer, No-Code Developer
- Timeline
- 2026 · 2 months
- Industry
- AI
- Platform
- Web, iOS, Android, desktop
An AI that doesn’t make you translate yourself first
Most AI products are built with one default user in mind, and everyone else adapts. You learn to prompt in a register that isn’t yours, you explain your slang, you correct the model when it gets your history, your names or your references wrong. SabiPT starts from the opposite assumption. It’s an AI assistant for Afro-Caribbean users (individuals, businesses and schools), meant to understand local languages, local context and local culture without being coached into it.
The product covers a lot of ground: conversation by text, voice and live mode, image creation and editing, deep search, and reading images or screenshots you upload. My work sat across UX and no-code build, which meant the distance between “this is how it should work” and “this is how it works” was short. I could design a flow in the morning and be testing it in a working build the same day.
Capability was never the hard part
The models could already do the work. The real problem was that a blank AI interface asks people to already know what they want and how to ask for it.
That gap gets wider when the culture you want to express isn’t the one the tool was trained to expect. Ask a general-purpose image model for “an Adire-inspired portrait” and you usually get a generic pattern with an African-sounding caption. The design challenge was to close the distance between intention and result: to let someone get to an output that actually looks and sounds like them, without needing to be a prompt engineer.
Give people the vocabulary, not a blank box
The image generator is where this thinking is most visible, so it’s the clearest way to explain the decisions.
- Choices before a promptMood, Environment and Illustration Style sit above the input. They turn “write a good prompt” into a few choices anyone can make, and they still combine with free text.
- Culture as the starting vocabularyAnkara, Adire, Kente and Ndebele are named, first-class presets, not a keyword you have to remember to add.
- Voice inside the prompt barSpeaking is often the most natural way to describe what you want, so it lives in the input rather than in a separate mode.
- Language is a top-level controlSwitching language is part of how people use the product, so it isn’t buried in settings.
- Honest about mistakesThe reminder that AI can be wrong stays in view, right where the decision to trust an answer is made.
Show the possibilities before asking for input. Example galleries do two jobs: they teach people what the tool can do, and they give them something to start from instead of an empty field. Picking a style and editing it is easier than writing one from nothing.
One assistant, many ways in. Typing, transcription, live voice and image upload all lead into the same conversation. The sidebar keeps the entry points that need their own space (New Chat, Upload Document, Image Generator), and everything else stays in the thread.
Honest about uncertainty. An assistant that claims to understand your context has to be careful about overconfidence. The persistent disclaimer and the product’s human-in-the-loop framing are there so trust is earned answer by answer rather than assumed.
Built for Africa, without the clichés
The visual language had to feel warm and rooted without treating “African” as decoration: no default kente borders, no pattern wallpaper. The interface itself stays calm: generous white space, a serif for headings that gives it some character, and a single orange accent for actions. The culture comes through in the content (the styles, the portraits, the language), so the outputs carry the identity and the interface gets out of their way.
Designing for context, not for a demographic
The easy version of this project would have been a generic AI app with Afrocentric styling. The more interesting version treats context as a product requirement: what the assistant offers first, what it assumes, which inputs it makes easy, and how it admits it might be wrong.
Working as both designer and no-code builder made those decisions concrete very quickly. When you can ship the change yourself, “should this be a preset or a setting?” stops being a debate and becomes a test.
Got a user problem?
Open to interesting product problems, collaborations and AI-driven ideas.
or write to obianujuohuegbe@gmail.com