What is HyphaHypha is the asset intelligence platform for commercial real estate professionals: lenders, private credit teams and asset managers. It turns fragmented documents (loan agreements, rent rolls, financial statements, appraisals) into a current, connected record of each asset, with every value traceable to its source.
I joined as the first designer, before there was any product design or brand, and today I lead design across product, brand and the systems behind them.
❋ Monthly Active Users 3x❋ Customer Orgs 3x❋ Grew team from 1 to 5❋ $50M seed launchBuilding design from zero meant figuring out a lot as I went. Here’s how I lead, what it taught me, and the systems that let a small team move fast.01 How I Lead
I lead from first principles, and I turn what we learn into systems, so the team doesn’t have to solve the same problem twice.
1. Start from the decision, not the feature.
Every screen exists to help someone make a decision. For a credit team, the decision behind almost everything is “can I rely on this number?” Framing it that way changed what we built. Instead of more dashboards, we invested in showing where each value came from, what changed, and who checked it
SYSTEMA trust model for AI output: source, change, confidence and review state on every value.
2. Measure the mess before you redesign it.
When the product started to feel inconsistent, the instinct was to redesign screens. We audited first. The problem wasn’t any single screen. It was the same thing being done many different ways: dozens of lists that opened details differently, and colours and shadows that had multiplied in code. That turned a vague “make it cleaner” into a short list of decisions.
A small set of core interaction patterns, starting with list-to-detail, and shared tokens that every new screen draws from.
3. Predictable beats clever.
In AI products it’s tempting to let the model decide everything. When we designed how drafts work, we tied them to a rule the team already uses instead of having AI guess. A rule people can explain is worth more than a smarter one they can’t.
A rule for AI features: automate the work, keep the logic legible
4. Fix the system that makes the screens.
If design and the shipped product keep drifting apart, more review won’t fix it. We made code the source of truth and changed who does what: engineers build the first working version, designers refine it directly in code, and AI checks keep it on system.
A design-to-code operating model
5. Put quality where the work happens.
As more teams needed design, I chose tools over gates: skills that let customer success, marketing and sales produce on-brand work themselves.
A rule for AI features: automate the work, keep the logic legible
02 What I learnt
Speed isn’t the value. Trust is.
I came in thinking AI's value was speed: extract in minutes what used to take hours. But the real cost for a credit team was never extraction. It was checking. If they don't trust a number, they redo the work and the speed is worth nothing. Now I design AI products around the moment someone decides to rely on the output, not the moment the output appears.Adoption is a design problem.
A product can be powerful and still go unused if people can't find their way into it. Our biggest gains didn't come from new features. They came from fixing structure, onboarding and search, so new customers reached value faster. Monthly active users grew about 3x as customer organizations went from 6 to 17. Now I treat "will people actually use this?" as a design question from day one.The real competitor is the old way of working.
Sitting with our largest customer as they moved off their legacy portfolio tool taught me that people don't compare you with other startups. They compare you with the spreadsheet or system they already know. Designing the switch (what carries over, what feels familiar, what's clearly better) matters as much as designing the destination.Small teams need deceptively more structure.
I expected process to slow a five-person team down. It did the opposite. Clear principles, regular reviews and QA meant fewer debates, less rework, and work that felt consistent even when we moved fast.AI moves design toward judgment.
When anyone can generate a screen, the scarce skills are deciding what should exist, building the systems that keep work coherent, and having the taste to know when something is right. That's where I think design teams will spend more of their time, and it's how I'm shaping ours.