# Peter Cuhalev, product designer & builder for technical platforms

> Full text of https://yukaii.com/, including every case study's full write-up, for AI assistants and answer engines. Contact: peter@yukaii.com · LinkedIn: https://www.linkedin.com/in/petercuhalev/ · Based in Slovenia (EU), remote with on-site visits.

*Open to full-time roles and project work · Remote*

## Product designer & builder for technical platforms.
Hi, I'm Peter. I shape how complex systems look and connect at a human level, design the experience people use, and ship with AI.

## From streaming data to EV payments, complex products made simple.
As a senior product designer & builder, I turn technical, complex systems into interfaces regular people can use, across big data and analytics, computer vision, EV charging and fintech, plus other enterprise projects.
Working as a bridge between product, business and technology shows me where the value is for users and how to present it. Knowing how the technology works means I know its limits, so solutions stay practical and ship sooner.
With AI, that means creating design systems AI can build from, and prototypes on the real codebase.
At the end of the day, I make things that get shipped, not things that stay in a drawer.
Companies I've worked with
- Road: EV Charging & Transactions
- Plainsight: AI Computer Vision
- Bottlenose: BI Analytics

## Selected work
Dashboards, apps and data platforms, across different companies.
- 01. Road, Dashboard
- 02. Road, Mobile App
- 03. Plainsight, Platform
- 04. Bottlenose, Platform
- 05. Alvenra, Side Project

### Designing Road’s platform, and the system AI builds from
*Road · Head of Product Design · EV Charging & Transactions · 2022–2026*
**Result: 1 week → 1 day to a first story draft**
Road runs pricing, payments and payouts for EV charging. I led product and visual design across its dashboard for end customers and its internal admin tooling, then wrote the design system AI builds from. Product and engineering now prototype with valid UI, and design spends its time on the experience and on checking we’re solving a real problem.
- One dashboard, three user groups — End customers, provider admins and system admins, designed as one product around pricing, payments and payouts.
- A design system AI builds from — Patterns, tokens, templates and blocks, so AI assembles on-pattern UI instead of inventing it.
- Adopted by product first, then engineering — Both teams now build from the same system, as Road moved to an AI-first way of building.
- Screen: Design system. AI context with definitions, values and templates.
- Screen: First pass from AI. Basic definitions, lacking domain knowledge.
- Screen: Full context. With complete domain knowledge, after testing.

#### Full case: Adapting AI to design inside Road's product
Time to a first story draft went from a week to a day, and the system was adopted by the whole product team first, then engineering.

##### Context
Road lets any EV charging station owner join its platform, then handles pricing, payments, and payouts. It also issues charge cards, offers white-label solutions, and runs its own roaming network. I owned product and visual design across the platform and all its products.

##### Approach
To create a design system AI could design from, these were the steps taken.
- Capture sources — Real Figma screens and the production app, with core patterns written down explicitly.
- Codify with AI — AI turned those definitions into its own product, design, and token files.
- Validate with testing — Existing stories ran through it, and outputs were compared between updates.
- Close the loop — Every failure became a rule; AI wrote its own do’s and don’ts.
- Constrain it like a human designer — Page templates and blocks gave AI real code to assemble from.

##### Screens
Example screens from the work created for this user story.
- Screen: Design system. AI context with definitions, values and templates.
- Screen: Template example. What AI assembles from, instead of inventing a page layout.
- Screen: First pass from AI. Basic definitions, lacking domain knowledge.
- Screen: Full context. With complete domain knowledge, after testing.

##### Reflections
AI accelerates; the foundation decides.
It’s only as good as the patterns and specifications underneath it.
80% comes fast. The last 19% is the real work.
And the final 1%? Nothing is ever truly done.

### Charging and paying from your phone, with or without a charge card
*Road · Senior Product Designer & Product Manager · EV Charging & Transactions · 2022–2026*
**Result: 350,000+ downloads across iOS and Android**
Road’s app lets EV drivers find a charger, start and stop a charge, pay, and manage their charge cards. When I joined it was a basic build from an external agency. I brought product and design in house, released the new app within six months, and designed paying in the app. It became a companion for drivers with a charge card, and a way for occasional drivers to pay without ordering one.
- Pay in the app — A full payment solution for EV charging, designed from scratch, so occasional drivers can charge without ordering a card.
- One app, 10+ brands — E-Flux by Road, plus white-label apps for more than ten customers, built from the same product.
- Built from the dashboard system — A mobile design system that kept the dashboard’s core components, tailored to the phone.
- Screen: Direct payment. Start a charge and pay in the app, by card, Apple Pay or Google Pay.
- Screen: Find a station. Sign up, add a card and start charging.
- Screen: Highway mode. The nearest fast stations in the direction you’re driving.

#### Full case: Charging and paying from your phone, with or without a charge card
350,000+ downloads across iOS and Android, as a companion to Road’s charge cards and a way for occasional drivers to pay without one.

##### Context
Road wanted to own the driver side of EV charging, next to its charge cards and dashboards, as a key part of its product stack. The existing app was a basic build from an external agency. I was brought in to start the project in house and deliver it, as E-Flux by Road and as white-label apps for more than ten customers.

##### Approach
To bring the app in house and make paying simple, these were the steps taken.
- Bring product and design in house — I took over from the agency, updated the app to current design standards, and kept building on top of the features it already had.
- Release the new app in six months — We started at the end of 2022 and released the new version about six months later. After that, development ran as a continuous track in weekly sprints.
- Keep payments as light as possible — Charging already has steps and statuses drivers need to learn. Paying in the app added a few more, so every extra step had to earn its place.
- Make prices transparent — Not every roaming partner sent us pricing, so a driver could get an unexpectedly high bill. Backend work and network agreements made the charging experience more dependable.
- Listen to drivers and customers — We tested internally with colleagues who drive EVs, ran a monthly survey to find pain points, and took white-label customers’ feedback into the roadmap.
- Scale white-label apps — I defined which parts of the app could be branded, worked with customers and developers on each published app, and wrote the process to move them to an external partner, so in-house developers could focus on the product.
- Keep it up to date — After launch we kept improving onboarding, payments, security and authentication, and released payments in all countries.

##### Screens
Example screens from the app, from finding a station to paying for a charge.
- Screen: Direct payment. Start a charge and pay in the app, by card, Apple Pay or Google Pay.
- Screen: Find a station. Sign up, add a card and start charging.
- Screen: Opening times. See when a station is open before you drive there.
- Screen: Highway mode. The nearest fast stations in the direction you’re driving.

##### Reflections
Good UX sometimes lives in the backend.
Price transparency needed data from roaming partners and network agreements, not only better screens.
Paying is part of charging, not a step after it.
Drivers already had steps and statuses to learn, so payment had to fit inside that flow rather than add to it.

### Computer vision without a machine learning expert
*Plainsight · Lead Senior Product & UX Designer, Product Owner · AI Computer Vision · 2017–2022*
**Result: Multiple tools → 1 flow from raw images to a deployed model in use**
Plainsight began as an engineering-built demo, split across separate tools with no way to deploy. I redesigned the platform and designed Sense 3.0 end to end, so non-technical users could train and deploy models. It brought the first paying customers.
- One flow instead of multiple separate tools — Every step visible up front: gather, annotate, train, deploy, run.
- One person can do it all — Larger teams could still give specialists access to their own steps.
- Templates by use case — Users saw their outcome first, with full configuration kept for ML experts.
- Screen: Annotate your inputs. The first of the tools a non-expert needs.
- Screen: Train on your own dataset. Accuracy, loss and training time against budget.
- Screen: Deploy where needed. Batch or streaming, image or video, each option explained in plain language.

#### Full case: Computer vision without a machine learning expert
Non-technical users could create, train, and deploy models to the cloud. First paying customers, and every development milestone met.

##### Context
Plainsight began as an engineering-built IoT demo. We saw a bigger opportunity in computer vision with regular cameras, which became Sense 3.0: an end-to-end platform from data gathering to running models on streaming video.

##### Approach
To turn an engineering-built demo into a product non-experts could use, these were the steps taken.
- Introduce an agile process — The process was engineering-led and product input was missing, so we introduced agile to close those gaps.
- Run pilots with customers — We were present in the IoT industry and validated the product through pilots with different customers.
- Listen to the strongest signal — One pilot surfaced computer vision and drew more interest than any before it. We listened to customers and pursued it, which turned out to be the right choice.
- Start with components and the platform — We designed the new product from a component system up, with the platform shaped first and features built on top of it.
- Default to common sense — Not knowing who each customer would be, we set common-sense machine learning defaults to ease adoption. Users could tweak them as they applied it to their own business case.
- User first, then technical — I designed simple options first and validated them with our ML experts, then filled in the technical requirements to capture all data and keep the journey seamless.
- Take it through to completion — In 2020 I stepped in as Product Owner, working closely with engineering to take the proof of concept to a finished product.

##### Screens
Example screens from Sense 3.0, following the flow from raw images to a deployed model.
- Screen: Gather your data. Sync status, progress and errors in one view.
- Screen: Annotate your inputs. The first of the tools a non-expert needs.
- Screen: Train on your own dataset. Accuracy, loss and training time against budget.
- Screen: Deploy where needed. Batch or streaming, image or video, each option explained in plain language.

##### Reflections
Strong technology doesn’t bring users on its own.
Adoption comes from making it user-friendly first.
Simple for everyone, even when the product is complex.
A foundation I still carry into my work today.

### Finding the trend in streaming data, before it grows
*Bottlenose · UX Designer · BI Analytics · 2011–2016*
**Result: Cool Vendor Gartner, 2016**
Bottlenose grew from social listening into a business intelligence platform that spotted trends early in social conversations, used by analysts and media teams at companies like Pepsi. I joined in 2011 as the first design hire, built its first pattern library and coded the front end, then took on more UX and UI direction as we moved from a consumer product to enterprise.
- Signal from the noise — Trend scores and change over time next to the data, so analysts could trust what the system surfaced.
- A pattern library from 2011 — Our own components and templates, so every new page and feature was built from the same parts.
- Alerts and reports — Keyword alerts and daily or weekly reports, so teams could follow their campaigns without watching the stream.
- Screen: Dashboards. The key numbers for a stream at a glance, with change over time.
- Screen: Compare. Two data series side by side, on one chart.
- Screen: Create a stream. Keywords and filters, with an estimate of how big the stream will be.

#### Full case: Finding the trend in streaming data, before it grows
Named a Gartner Cool Vendor in 2016, as Bottlenose moved from a consumer product to an enterprise platform.

##### Context
Bottlenose started as a social listening platform for finding trends in social networks, and grew into a business intelligence platform that correlated business data with social conversations. In Nerve Center, users built streams around their own keywords to surface small organic content before it grew into a bigger trend, so companies could ride that wave. It was used by social analysts and media content teams, including Pepsi.

##### Approach
To make streaming data useful for analysts, these were the steps taken.
- Bring the new design into code — I joined after the rebrand as the first design hire, coding the latest designs in HTML and CSS to roll out the updated product.
- Work in a tight loop — As the only designer, I worked directly with the CPO, CTO and engineers, so specs, builds and feedback moved quickly.
- Check with our analysts — We reimagined features internally, then asked our analysts if what we surfaced matched what they knew from the market. That showed the gaps and kept the roadmap on the right problems.
- Follow the market to enterprise — We were still finding product-market fit, and clients didn’t always mean paying clients. We pivoted quickly, and an enterprise solution for bigger media companies brought the paying customers.
- Validate with real use — Customer feedback, internal testing and analytics showed which parts of the app were used most and which were undervalued.
- Grow into UX direction — Alongside the front end, I took on more UX and UI direction together with our CPO as the team and company grew.

##### Screens
Example screens from Nerve Center, from setting up a stream to analysing it.
- Screen: Dashboards. The key numbers for a stream at a glance, with change over time.
- Screen: Analyze. Entities ranked by volume, next to their metrics over time.
- Screen: Compare. Two data series side by side, on one chart.
- Screen: Create a stream. Keywords and filters, with an estimate of how big the stream will be.

##### Reflections
Trust comes from showing the working.
A score next to every trend, with the data behind it, gave analysts confidence in what the system surfaced.
Familiar tools beat clever ones.
Our Sonar network map looked great, but analysts validated in a line chart and a table, so those stayed at the core.

### A sound meditation app, from idea to live product with AI
*Alvenra · Concept, product, design and front-end · Side project*
**Result: Idea → live designed and built end to end with AI**
Long-form guided sound meditation, with intention setting and integration, isn’t available digitally. I defined the product, flow and visual style first, then built it with AI, well past the first prototype.
- Defined before building — Idea, flow and visual style written first, so AI worked from a clear definition.
- Past the first prototype — A cymatics visualizer refined over many rounds to production quality.
- Fixed what breaks the experience — Audio now keeps playing with the phone locked.
- Screen: During. One control, nothing else.
- Screen: Before. Setting the intention.
- Screen: After. Private journaling.

#### Full case: A sound meditation app, from idea to live product with AI
A live product, from concept to deployment, designed and built with AI.
Open the live preview

##### Context
I facilitate sound meditations, and wanted to explore a digital way to offer them to participants. Long-form guided sound meditation, with intention setting and integration, isn’t common in meditation apps today.

##### Approach
To take it from idea to a live product with AI, these were the steps taken.
- Start from the in-person session — A live session has a shape: setting an intention, the sound journey itself, then time to integrate. I kept that shape as the backbone of the app, so every session moves through before, during and after.
- Write the definition first — Before any code, I wrote down the idea, the flow and the visual style. That gave AI a clear definition to build from, instead of leaving it to guess.
- Let the session take the stage — During a session there is only one control. Everything else steps back, so the sound and the visual carry the experience.
- Push the visualizer past the first prototype — AI’s first version worked, but felt generic. The cymatics visualizer went through many rounds of refinement until it matched what I had in mind.
- Keep integration optional and private — After the session, journaling prompts help people reflect on their intention. They are optional, and what people write stays private.
- Fix what breaks the experience — Audio stopped when the phone locked, which breaks a long session. Playback now continues with the screen off.
- Ship it live — Designed and built end to end with AI, from concept to a deployed live preview anyone can open.

##### Screens
Example screens from the live app, before, during and after a session.
- Screen: Before. The intention and how to take part.
- Screen: During. The visualizer, a progress ring, and one control.
- Screen: After. Optional, private journaling prompts.

##### Reflections
Don’t start with AI; start with yourself.
A first pass of your own gives AI a far stronger definition.
AI enhances craft; it doesn’t replace it.
The polish comes from your own eyes.

## Designing with AI
AI is part of every working day for me now, from product research and first ideas to prototypes and checking UI before it ships. Here’s how I use it, what I’ve learned, and where I think it’s heading.
Daily tools: Claude, Claude Code, Figma AI agents

### Day to day
- Specs before prompts — I start from my own understanding: specs and definitions, or a story the team brings in. That sets the parameters before AI touches anything.
- AI makes the first draft — Claude and Claude Code turn that into a first draft and a few explorations, so I can see options instead of imagining them. On bigger projects, Figma is still where I explore design-only features and set up the system.
- I choose and refine — I pick the direction that works, prompt specific changes until it fits the system, then tweak by hand on top of what AI made.
- Checked before anyone sees it — Output runs through design specs and review checks first, so my review can focus on value, not inconsistencies. Then it goes to a client, a customer or production.

### What I’ve learned
- AI gets you to a solid solution much faster — Tweaking and validating still take about the same time, if you don’t want to ship AI slop.
- Sometimes my hands are faster — When I know exactly what I want, making it myself beats many rounds of prompts. It depends on the feature.
- It’s only as good as our definitions — I first expected finished designs with nothing to tweak. We keep adapting as people, and so do the definitions AI works from.

### Where it’s heading
- Design moves toward judgement — More product thinking and deciding what’s good, plus the craft and visual cues that keep a product unique and human.
- Checks built into the system — Product design specs, review and polish steps that every AI output passes through, so mistakes surface early and gaps close sooner.

## Ways to work together
I join teams full-time, working remotely with on-site visits.
I’m also available for project work.

### Product designer & builder, joining your team.
Product thinking, UX and visual design across your platform, with AI-assisted prototyping on your real codebase. Working alongside product and engineering every day.
- Availability: open to new roles; see https://yukaii.com/ for current availability
- Location: Remote, with on-site visits. Based in Slovenia, EU

### Project work
- Product and UX design — A feature, a flow or a whole platform, from definition and user journeys to the visual details people feel.
- Design system and UX review — An assessment of your system, patterns and flows, with a clear list of what to fix first.
- AI-ready design systems and process — Structure your system and workflow so AI builds from it accurately, without losing design quality.

## Experience
Designing products across streaming data, computer vision and EV charging, from print and web to co-founding a startup and leading product design.
- 2022–2026 — Road — EV Charging & Transactions — Head of Product Design · Senior Product Designer · Product Manager (Mobile App)
- 2017–2022 — Plainsight — AI Computer Vision — Lead Senior Product & UX Designer · Product Owner
- 2011–2016 — Bottlenose — BI Analytics — UX Designer
- 2009–2012 — Preona — Web, Browser Plugin, iOS App — Co-founder and CPO
- 2005–2008 — Various companies — Print and web — Graphic and Web Designer

## Let's talk about your product.
Hiring for a product design position, or want a second pair of eyes on your design system?
Email is the fastest way to reach me:
peter@yukaii.com

## Contact

- Email: peter@yukaii.com
- LinkedIn: https://www.linkedin.com/in/petercuhalev/
- Website: https://yukaii.com/