気
SPIRIT (KI) — HUMAN INTENT & RIGHT TO CORRECT
The user’s intent and their ability to intervene must be present. Without a real, immediate right to edit, override, or regenerate the output, trust collapses the moment the model is wrong.
“The hardest thing to design is not the interface. It is the relationship between a person and a system they cannot fully predict.”
DanielCanabrava
Every product decision changes someone's life. Not metaphorically — a failed feature is someone's job harder, someone's trust broken, or people losing work. That weight has never been separate from the work for me.
In 2017, at Brazil's Federal Court of Accounts, I watched a skeptical senior director read an AI-generated draft of a quasi-legal document twice — then blush — before saying it would turn months of work into hours. The platform reduced instruction-writing time by 63%. What stayed with me was the moment: how a guarded expert decided the output was reliable enough to act on, without being asked to take it on faith.
That question has organised my career: how do you design the relationship between a person and a system they cannot fully verify? I have ADHD. My brain does not move in linear steps; it pulls analogies across Kendo, audit bureaucracy, behavioral economics, and GenUI by default. That is not a technique I learned — it is how my mind operates. Empathy and the drive to earn genuine trust are not values I added to my practice. They are the Kihon I drill every day. 13 years on the craft. I turn non-deterministic systems into experiences people can rely on.
In 2017, while usability-testing an AI-powered auditing platform at the Brazilian Federal Court of Accounts, I watched a senior Secretariat Director — known for being extremely demanding and skeptical — read an AI-generated draft of a complex legal document.
He read it once. He read it again. Then he blushed.
He turned to the team and said it would turn work that normally took months into a matter of hours.
That single moment reorganized my entire career.
I had already been designing for high-stakes, non-deterministic systems for years before “AI trust” became a category. The platform ultimately delivered a measured 63% reduction in instruction-writing time and won a national award for technology in the public sector. But what stayed with me was not the metric — it was the moment a guarded expert decided the output was reliable enough to act on, without being asked to take it on faith.
That question has guided everything since: How do you design the relationship between a person and a system whose output they cannot fully verify?
He read it once. He read it again. Then he blushed.
He turned to the team and said it would turn work that normally took months into a matter of hours.
That single moment reorganized my entire career.
I had already been designing for high-stakes, non-deterministic systems for years before “AI trust” became a category. The platform ultimately delivered a measured 63% reduction in instruction-writing time and won a national award for technology in the public sector. But what stayed with me was not the metric — it was the moment a guarded expert decided the output was reliable enough to act on, without being asked to take it on faith.
That question has guided everything since: How do you design the relationship between a person and a system whose output they cannot fully verify?
01 — What I Do
Trust in AI Systems
I design the layer where people decide whether they can rely on an intelligent system. Not by adding trust signals at the end, but by reshaping the inputs the model sees upstream and giving users a real, immediate right to correct what comes out.
AI Product Design
When behavior is probabilistic and generated at runtime, the interface's real work happens in the uncertain states. I design for partial results, low confidence, and outright failure first — because that is where users either learn the system is worthy of trust or learn never to trust it again.
Analogical & Visual Thinking
I don't invent patterns for new problems. I find the ones users already know from the rest of their lives and map them onto the AI interaction. The same input-shaping logic that made a 2017 government audit system reviewable by skeptical directors is the logic I use on LLM interfaces today. Users don't live in your product.
Close to Code
Trust only counts if it ships. I work in sketches and high-fidelity prototypes in TypeScript and React, and I stay through implementation and production review. The mechanisms I design are only real once they survive actual users and actual engineers.
02 — Parallel Trust Design
I don’t follow linear design processes. Understanding the real problem and constructing the solution develop together, constantly reshaping each other.
I run two tracks in parallel:
I run two tracks in parallel:
Track A — Sensemaking
Visual pattern-finding is the hub. From there I run deep detective immersion into the data and the user’s actual situation, draw analogical connections across domains (Jacob’s Law), do philosophical framing, and stress-test ideas through team debate. The goal is to isolate the actual uncertainty the user cannot resolve on their own — before any interface is drawn.
Track B — Trust-Design Toolkit
With the real problem clarified, I focus on the two layers that actually produce trust in non-deterministic systems:
- • Reshape the inputs the model sees, so its outputs become more predictable and legible.
- • Give users the explicit, low-friction right to correct what the system produces.
“I earn AI trust at two layers: I reshape the inputs the model sees so its outputs are more predictable, and I give users the right to correct what it produces.”
03 — Strategic Alignment Spine
A viable product in uncertain systems requires three vectors to land together. In Kendo this is called ki-ken-tai-ichi — spirit, sword, and body as one. A strike only counts when intent, action, and form arrive in the same instant. I have trained to second dan for seven years. This is not a borrowed metaphor. It is a principle I have drilled physically, and it shapes how I design the relationship between a person and a system they cannot fully predict.
剣
SWORD (KEN) — PROCESSING TELEMETRY & LEGIBILITY
The system’s working must be made visible as it happens — streaming states, citations, reasoning traces, and generative components forming in real time — so people can follow what the system is actually doing instead of having to guess.
体
BODY (TAI) — INPUT CONSTRAINTS & SHAPING
Clean, structured inputs are the foundation. At the 2017 TCU audit platform, dozens of inconsistent legacy templates were consolidated into one patterned system so the model could read them reliably. Structured inputs produce outputs worth correcting.
63
63%
reduction in instruction-writing time on a shipped government AI platform
40
~40%
revenue growth contribution from a core product redesign
Beyond the numbers: strategic prototypes that changed company direction, internal AI tools that accelerated team velocity, and consistent mentorship that raised the floor of entire design teams.
05 — Experience
Nortal · BTG Group — Capstone Technology
UX Engineer
Nortal · Markerr
Senior UX Designer — LLM & AI Products
AI Solutions
Senior UX Designer
Ília Digital
UX Design Lead
Capgemini
Senior UX Designer
Configr
Senior UX Designer
06 — Skills & Tools
AI Product Design
LLM & GenUI Interfaces
Agentic Workflows
Trust & Uncertainty Design
Behavioural Design
Design Systems
Figma
Framer
Webflow
React & TypeScriptStorybook
Claude Code
Cursor
Grok
Gemini
OpenAI
Perplexity
GitHubResearch & Analytics
Kendo (nidan)
07 — Design Principles & Lived Discipline
Parallel Trust Design
I reject linear design processes. The understanding of the user’s problem and the structure of the technical solution co-evolve. Visual pattern-finding sits at the center. Two tracks run in parallel: reshaping inputs for predictability, making the system’s behavior legible, and giving the human an immediate right to correct.
Kihon as Foundation
Empathy and the drive to earn genuine trust are not values I added to my practice. They are the foundation it is built on. Everything else — the input architecture, the behavioral translation, the code proximity — comes after that. These are drilled fundamentals, not poster slogans.
Craft Lives in Failure States
The states where the model is uncertain, partial, or wrong are not edge cases. They are where lasting trust (or lasting distrust) is formed. I design those states first.
Ship Close to Code
I build high-fidelity interactive prototypes in HTML, CSS, TypeScript, and React. I stay through implementation and review production against intent. Static frames over the wall are not how trust ships.
The Only I
Only I turn non-deterministic systems into experiences people can rely on — because my brain works as an outlier by default. I have ADHD. I do not think in linear steps or follow methodologies. I think in analogies and visual patterns. My cognition pulls connections across domains that others treat as unrelated — Kendo, behavioral economics, semiotics, audit bureaucracy, GenUI. That is not a technique I learned; it is how my mind operates. I craft with care and time. I am held accountable for everything that ships. 13 years on the craft. Kihon. Excellence in the fundamentals, drilled until they become instinct.
“A strong balance between design expertise and development collaboration… works effectively with engineering teams to ensure smooth implementation… willingness to challenge conventional approaches.”
“Thinks outside the box while keeping focus on what’s most important… a fresh take over already established conventions.”
“An inspiring leader and an incredible mentor… working alongside someone as talented and dedicated as Daniel left a positive mark on my professional development.”
I’ve spent the last several years treating non-deterministic systems as a design problem first. The next phase feels like it needs a different quality of attention.
I’m increasingly interested in what Zen calls mushin (no-mind) and shoshin (beginner’s mind) — the ability to act with full capability but without the friction of self-conscious effort, and the capacity to meet each new problem as if for the first time.
In practice this points toward interfaces that feel less like “operating” an intelligent system and more like a fluent, low-friction exchange. Less dashboard, more practiced conversation.
I’m increasingly interested in what Zen calls mushin (no-mind) and shoshin (beginner’s mind) — the ability to act with full capability but without the friction of self-conscious effort, and the capacity to meet each new problem as if for the first time.
In practice this points toward interfaces that feel less like “operating” an intelligent system and more like a fluent, low-friction exchange. Less dashboard, more practiced conversation.
Beyond the work.
7+ years. The source of ki-ken-tai-ichi and zanshin.
World Embassy Cup
Ride
The world
Another form of creation
Learning ink painting
Learning Japanese calligraphy
Let's build something.
I'm currently open to senior product design and design-lead roles at growth-stage companies building AI-native products where the quality of the interaction layer is a real constraint, not an afterthought. If you're working on systems where people have to act on outputs they cannot fully verify, and you care about the craft of making those systems legible, correctable, and worthy of trust — I'd like to hear about it.