AI and fintech startups we build with

Holdex is a Hong Kong design and technology studio. We help founders ship AI and fintech products to production, including regulated digital-asset products.

We say no to most teams that approach us, and the idea is rarely the reason. We say yes when a product needs to reach production and we can be the team that gets it there. This page explains how we decide, so you can judge the fit before you book a call.

Who do we build with?

AI is our main focus. Some teams we work with are building a product where the AI is the product. Others are adding AI to an industry that hasn't used it yet, such as insurance, construction, logistics, or medicine. In our experience, that second group is further along on the model than they expect, and further behind on everything around it.

Fintech is our second focus: payments, lending, wealth, neobanks, the systems behind them, including compliance and regtech, and digital assets for institutions. What we build covers both areas.

We don't ask which round you have raised. We ask whether your product has to ship in the next few months.

Many founders now arrive with something that already works, often a prototype built with AI coding tools. What's missing is everything that turns it into a real product: login, billing, data, reliability, and the records a bank or a large company will ask to see.

What do we look for?

A problem people would miss

Our first question is simple: if the problem disappeared tomorrow, would anyone notice? If you can't answer in one sentence, the product needs more thought before it needs engineers.

A team that has shipped before

We look for founders who have finished something before, in any setting, and who recovered when part of it went wrong. Knowing the industry helps. Being able to cut features to hit a real deadline helps more. More on this in why credentials answer the wrong question.

Something real to start from

A prototype, a pilot, a waiting list, or one user who has agreed to try it. Our best projects start from something that already exists and that a real user has touched, even briefly.

You don't need to write a requirements document. We turn what you tell us into a spec, and anything it can't answer comes back to you as a decision.

When do we say no?

AI added only because it's expected. If a feature exists to look good in an investor update, your team ends up maintaining code no user asked for. If the product still works with the AI removed, the AI was never the product.

A demo that will stay a demo. If you want a prototype for a pitch and nothing after it, we're the wrong team. Most of the work is making a product hold up with real users, the right access controls, and audits, and that's the work we do. Why an AI demo is not a product explains the gap.

Big plans with no first step. Big plans are welcome. The founders we work with can also say what has to be true in 90 days, and what they will cut to get there.

What happens on the first call?

We talk about your business, not your tech stack: what you want to achieve, what's in the way, and what success looks like a year from now. We keep the conversation there until the problem is clear, because the technical decisions follow from it.

If we're not the right fit, we tell you on that call, and where we can, we suggest someone who is. If we are, the next call is about what to build first and in what order.

What happens after we start?

The senior people you meet before we start are the people who build your product. They work inside your team from the first week, and they stay after launch. A Holdex team has run Clearpool's product and engineering since 2021. How we work with founders covers team size and the first weeks.

Founders we work with also get access to our network and to startup perks that stretch your tooling budget in the first year.

You can check how we work before you commit. Our engineering rules are public on GitHub, in holdex/developers. Every change goes through a pull request your team can read, and a person decides what gets merged.