UXBOX Company LTD
Vietnam’s first AI-powered platform for navigating legal and administrative procedures. Built for people, optimised for AI agents. Built to surface in AI answers and turn discovery into leads for registered professionals.
Services
AI features, past the chat box
Two years ago an AI feature was the product. Now it is the baseline: people assume the tedious parts are handled, and a product that still asks them to do it by hand feels dated. That rarely means adding a chat panel. The features that earn their place are usually invisible: a field that fills itself, a list already in the right order, a summary above a long thread.
What the work covers
Where a model actually helps
We start with what people retype, wait for, or contact support about. Those are the candidates. A chat box is rarely one of them.
Interfaces that are not chat
Suggestions in place, defaults already filled, drafts to accept or reject, a summary above the detail. The model does the work, and the product stays a product rather than becoming a conversation.
Designing for the wrong answer
Confidence shown honestly, an obvious way to undo, and a path back to doing it by hand. Trust is difficult to recover once lost.
Latency as a design problem
Models are slow and uneven. What the screen does with those seconds is the difference between a product that feels fast and one that feels broken.
Limits, costs and what gets sent
What leaves the product, what each action costs, and what happens at the rate limit. These are decided in design rather than discovered in production.
Evaluated before it ships
Real inputs from your data, an agreed idea of what a good answer looks like, and sessions with the people who will live with the result.
How it runs
We start from the job, not the model
What is someone trying to do, and where is it slow or repetitive? Then whether a model is the best answer. Sometimes it is a better default or a saved filter, and we will say so.
Prototyped on your data
Sample data makes everything look convincing. We test on your real data, where hallucination, latency and edge cases actually appear.
Shipped small
One feature, in front of real users, with the fallbacks in place, before it becomes the basis for four more on the roadmap.
The failure path is designed too
The wrong answer, the empty answer, the slow answer and the refusal, given the same attention as the case where it works.
What it costs
Fractional is $4.2K a month, Full-time is $7.7K a month, and fixed-price projects start from $30K.
Selected work
See all work
GoTeach Collective
GoTeach.vn is Vietnam’s first platform connecting private tutors and coaches with students. In just 3 months, we helped them validate their idea, design their brand, build the platform, and launch.
WIPNetch Digital
Invoicing.mu aims to become Mauritius’s leading e-invoicing SaaS in terms of user experience and seamless onboarding.
Other services
Questions we get
More questions about working with usDoes our product need AI features?
Only where they remove work a customer is doing by hand, and increasingly customers expect that, because the products they use all day have set the baseline. If there is no such job in your product, adding a model costs you more than it returns.
Does this mean building a chatbot?
Usually not. Chat is one interface, and it is the one that asks the customer to do the thinking: work out what to ask, phrase it, judge the answer. Most of the value is in features that do not look like AI at all: a draft ready to edit, a list already ordered, a form that arrives filled in.
What about hallucination, and trust?
Design for it. Show what the answer is based on, make the confidence honest, keep undo one click away, and leave the manual path in place. A feature that is right most of the time and unrecoverable when it is wrong will be turned off.
Our data is sensitive.
Then what leaves the product is a design decision rather than an implementation detail. We work within your constraints, covering what is sent, what is retained and what stays on your side, and the interface states this where a customer would reasonably want to know.
Can you build it, or only design it?
We build MVPs ourselves with AI-assisted workflows, and keep a vetted network of engineers for production work. On an existing product we design for your team and stay available while they build it.
How much does this cost?
AI work is part of the engagement rather than a separate line: Fractional is $4.2K a month, Full-time is $7.7K a month, and fixed-price projects start from $30K. The pricing page states all three in your own currency.
Work with us