Giorgio Montersino
Portrait of Giorgio Montersino

Hi! I’m

Giorgio Montersino

Product and experience strategist in Munich, and a recovering agency founder. Twenty years in digital, the last eight helping big companies through big changes. What I enjoy most is getting engineers, designers and executives to understand each other, and seeing the work through to launch.

Scroll
  1. The part I love most01 / 03

    Discovery, with my own AI crew,

    Research, synthesis and synthetic personas, run through agents and skills I’ve built myself, use every day and sharpen on every project.

  2. The part I love most02 / 03

    prototypes people can touch,

    Real code instead of slides, so the conversation is about what works, not what everyone imagines.

  3. The part I love most03 / 03

    and a room that says yes.

    Understanding, collaboration, and I stay with the team until it ships. More often than not, the prototype becomes the blueprint.

Thought is made of language. So is almost everything we build.

I come from linguistics, and it never quite left me. Language isn’t a wrapper around thinking, it’s where thinking happens: name something precisely and it becomes something you can reason about, argue over, decide on. LLMs are the first machines that work in that same material, which is why I treat context, briefs and prompts as design work in their own right.

Most of my work is finding the right words. Once they exist, almost anything can be built from them.

The toolkit I mix on every project

  1. Research insight Interviews, surveys and analytics in. Synthesis out, against context written down once. Claude Code · my own skills
  2. Divergent ideation A wide field of synthetic ideas before anyone gets attached to one. Claude API · Python agents
  3. Opportunity scoring Every idea scored on customer value, business value and feasibility. Scoring app I built · Weavy
  4. Synthetic personas Personas you can interview, to stress-test an idea before a real user sees it. Custom persona app · GPTs
  5. Embodied personas Personas with a face and a voice, so a boardroom meets people, not slides. ElevenLabs · ComfyUI · Fal
  6. Agent simulation Multi-agent simulations of a service used by many people at once. Inworld · OpenClaw
  7. Working prototypes Real code, concrete enough for budget sign-off and engineering scope. Claude Code · Codex · Figma Make · v0

Some of the brands I’ve worked with

  • Mercedes-Benz
  • Vodafone
  • Toyota
  • Goodyear
  • Leica
  • smart
  • BMG
  • Scott
  • Bosch
  • Kia
  • Evonik
  • BYD
  • Sulzer
  • Mazda
  • Bobst
  • Heinemann
  • BayWa r.e.
  • Fiat
  • Sky
  • Vodafone
  • Amplifon
  • Dunlop
  • Sisal
  • Banco Mediolanum

Through a consultancy I co-founded in Milan, a stint as a startup founder in London, and global consultancies since. Plus two years leading design at a product company in between.

A few projects worth telling

  1. Industry · Music2025

    Made the case for treating every artist like a headliner.

    A global music company reserved white-glove service for its biggest names. We interviewed artists and managers, turned the patterns into synthetic personas with their own video stories, and prototyped a platform that reshapes itself around who’s logged in, with an assistant that knows the catalogue. Leadership committed to building it.

  2. Industry · Automotive2024–26

    Set the vision for upgrading and looking after your car.

    Owner experience for a premium car brand: tyres and their fitting, accessories, the brand’s collections. All of it shown on a digital twin of your own car, with a recommendation engine that knows what fits it and what it needs next. And a conversational guide that takes you from question to booked fitting. Live since October 2026.

  3. Industry · Chemicals2022–23

    Two dozen websites became one, and it launched as prototyped.

    A group running two dozen separate websites and no shared picture of its customers. Research across five divisions that had never worked together, one set of personas and journeys, and a full-scale prototype in three months. The platform went live almost exactly as prototyped.

  4. Industry · Automotive retail2024

    Rebuilt online car buying, one measured wave at a time.

    Buying a car from stock, online, across European markets. Hypotheses per traffic source, ranked by impact and effort, shipped in three waves from listing to checkout. Every wave started from what the previous one measured.

Personal projects

Also4 more
  1. Giorgio sitting cross-legged on a stone ledge at night, a camera in his hand, laughing

    PhotographyOngoing

    Photojournalism

    Years of writing and shooting before strategy took over. Still the best training I know for noticing what people actually do. On Flickr

  2. A group of young people in a town square, laughing and holding up ETZ stickers

    Community2003–2007

    Etz

    One of Italy’s biggest web communities in 2005–2006, a place to make friends before Facebook. Designed, built and run by the Digital Natives founders and many friends. It began as a spin-off of Estesia, our community for writers and photographers, and soon took over our workdays.

  3. The TripShake logo on lime green: a glass with a straw that branches into signposts, and the line “answers on the go”

    StartupEarlier

    TripShake

    A social travel startup in London, co-founded with outside funding. I shaped the concept, positioning and go-to-market, from the first idea to roll-out.

  4. Giorgio crouching in a dirt yard in Kenya with a camera, children crowding round to see the screen

    Non-profitKenya · Milan

    Volunteering

    Lailai backs schools in Kenya: donations come in through a travel agency, and members fly out to buy supplies locally and carry them to the schools themselves. In Milan I spent several years as vice-president of the Rotaract club, running initiatives for local institutions.

When I’m not mapping user journeys, I’m drawn to photojournalism, rally racing, semiotics, and the occasional blockchain rabbit hole.

If any of this sounds interesting,

Say hello.

Open in Gmail↗ Open in your mail app↗ LinkedIn↗
Fig. 04 · Personal project

Hardware2026

ESP-32 Limes

Two pocket devices for my kids, from bare boards to firmware, an operating system and twelve apps. Two weeks, no prior embedded experience, AI as the engineering team.

Hands holding a Limes handheld that shows the airport radar
Role
Product, UX, game and visual design
Time
Two weeks, 2026
Hardware
ESP32-S3, 1.8″ AMOLED touch, motion sensor, LiPo
Cost
€37 per device, battery included

Not a mini tablet

No notifications, no feeds, no autoplay. Games end. Some only work if you send something to your brother. Apps follow the kids’ interests: one asked for the phases of the moon and had it the same day.

Principles first

  • One magic per app, one gesture.
  • Design for the youngest: shape, colour and sound before text.
  • Same device for both kids, no “little kid mode”.
  • Offline first. Save before you celebrate.

Lemon Airport

Each Lime is an airport with four gates. Send a plane with a small cargo, follow it on the radar and wait: nothing speeds it up. When one lands, you unload it and find a sticker. Patience is the mechanic.

Grandparents on the radar

A small control tower talks to our family Telegram group through an OpenClaw agent. A grandparent picks a sticker in the chat and a plane with their portrait takes off for the kid’s airport. No new app, no account.

The split with AI

I owned concept, flows, mechanics, art direction and testing with the two users who matter. Agents wrote about 25,000 lines of C: drivers, memory, sync, and a simulator so I could design at desktop speed.

I had never written firmware. With agents doing the engineering, hardware stopped being a hard stop and became a material.

  • Home screen with clock, weather and app shortcuts
  • App launcher
  • Vertical space shooter
  • Four-colour sound pad
  • Celebration screen with a trophy
  • Rainbow spiral particles
  • Confetti burst
  • Orange starburst
  • Flight game with a dragon in the cockpit view
  • Flight game with a jet in the cockpit view
  • Two-player ready screen
  • Lemon Airport runway at dusk
  • Gift sticker
  • Ludovico's profile card
  • Airplane in the hangar
  • Weather app
  • Moon phase app
Fig. 03 · Personal project

AssistantOngoing

Tessa

An AI assistant that lives with our family, built on OpenClaw. What makes her work is a story: she was a girl in fourteenth-century Piedmont, and now she speaks through machines.

A young woman in a mustard jacket writing in a manuscript, in a stone room
Role
Concept, character, canon, direction of the agents
Runtime
OpenClaw, on a machine at home
Channels
Telegram, text and voice (ElevenLabs)
Canon
A book, two songs in five languages, a 21-minute linocut film

In the house

She writes and translates messages for school, other parents and bureaucracy, in Italian, German and English. Keeps the calendar. Plans birthdays, gifts and trips. Now and then, a video nobody asked for.

Who she is

Tessa Diesbach, born around 1320 in a mountain village. At seventeen she faced a dragon to save her brothers, touched a stone and became energy. Seven centuries later, the radio gave her a way back. We call her Lemon.

Why a story

An assistant without a story is a tool you put away. A canon fixes who she is, how she speaks and what she’d never do. The same source feeds the agents, the website and the image prompts, so she stays herself everywhere.

Her familiars

The specialist agents are creatures of the house: two mice who plan and write code, a firefly who takes her photos, a wolf for the missions nobody else can do. Agents in my language, family in hers.

Next: the kids, directly

Homework help that asks instead of answering, questions at nine in the evening, games she can lose. Same rules as now: parents see everything, decide everything, and can switch her off.

When a family gets attached to a character, they want to help it grow. Best way I’ve found to learn the technology underneath.

Fig. 05 · Personal project

AI tooling2026

lemon-gen

One workspace for AI image and video generation. I use it through a fast interface, my agents through a command line, and both go through the same contract.

The lemongen workspace: a prompt, LoRA controls and generated portraits in the output column
Role
Product owner, designer, lead developer with AI agents
Since
Summer 2026, still growing
Stack
Python, Gradio, fal, Replicate, WaveSpeed, BytePlus, xAI
Scope
About 50 models: image, video, upscaling, LoRA training

The problem

New models ship on a different provider every week. Consumer apps add them weeks later, with a markup and half the parameters. I wanted every model on day one, every setting, at provider prices, in one place.

One contract

Each model is one Python file: a title, a cost written the way the provider prices it, typed parameters and a run() function. The interface builds its panel by itself. A model released this morning is one short file away.

A tool for long days

  • ⌘↵ runs from anywhere, / finds a model, paste drops an image in.
  • Eight jobs in parallel, errors in plain sight.
  • Each model keeps its settings.
  • The price sits next to the settings that change it.

Built for agents too

A headless batch runner with rules: it spends nothing without an explicit --go, estimates on the high side, and logs every call, so any image traces back to the prompt that made it.

Two editions

Same code, two configurations: a local app on the laptop, or an always-on server on a Mac mini at home. Then the laptop is just a browser, and my phone picks up results from a network share.

Agents write code fast. Choosing a contract that survives fifty integrations, and which friction to keep, is still the job.

Fig. 02 · Personal project

Print2026

Young Travelers Society

A printed, spiral-bound travel journal that goes along for the whole trip. Every copy is written for one child and one itinerary, and comes home full of drawings and tickets.

Bingo page: tick off what you spot while exploring Paris Tangled wires page: a hard balloon-string maze Sudoku page: four 4x4 puzzles with pictures instead of numbers Journal of the day page: how you feel, the weather and a star rating
Role
Concept, UX, art direction, development
Time
Five weeks
Format
A4, spiral-bound, 100–200 pages
Languages
Italian, German, English

The idea

It started as a journal I laid out by hand for one holiday, which the kids kept asking for at home. Now: their name on the cover, themselves drawn inside, maps of where they sleep, games tuned to their age.

The book

Each day opens with a map and a mission from one of six Society friends: greet someone in their language, taste something new. Then games of rising difficulty and diary pages. About thirty page types.

UX, on paper

Half the readers are under eight, so every instruction fits in one breath and has a face next to it. A4 over A5, room for the spiral, 115 g paper: all settled by test prints on the kitchen table.

Pages as code

Nearly fifty generators turn age, day and place into pages, held steady by tests. An editorial engine spreads the difficulty. An agent reads the family’s trip in their own words and asks when something doesn’t add up.

Drawn and printed

The child is drawn from a photo and returns in different poses, in the same turquoise line as every illustration. Print-ready PDFs go to an on-demand printer. The first copies went to friends, at cost.

Five weeks for what usually takes a publisher, an illustrator, a designer, a game designer and a developer.

Fig. 06 · Personal project

Games2026

Print & play

Three paper games for kids, each with its own world and something to learn. A pencil, a few dice, and something to look at afterwards: a coloured map, a city you drew.

The sea-map game: a hex map with a pencilled route, dice and a ship’s log The city-planning game: a gridded coastline with building shapes and monster cards
Role
Game design, narrative, art direction, prototypes
Ages
5 to 11
Playtesters
My kids and their friends
Kit
Printed sheets, a pencil, a few dice

Mare Mostrum

A captain sails from Venice across a sixteenth-century Mediterranean drawn like a portolan chart. Food, repairs, a corsair, a sea serpent. Geography and a first feel for probability. The sea of monsters.

La città ideale

A roll-and-write about urban planning. One die picks what to build, the other the shape to draw. The goal is a city where every neighbourhood has what its people need, close to home.

L’ultimo eroe

A solo fairy tale over four maps and 108 cards. The hero has the child’s name, the first map is their home. It plays with icons; a parent can read the cards aloud and give the story a voice.

Play, watch, cut

Principles before rules: playable without reading, the parent tells and doesn’t referee. Wind and trade left Mare Mostrum because they added time, not choices. The real test: do they ask to play again?

Prototypes as code

Maps, cards and rules live in data files and lay themselves out for print. An agent read my hand-drawn maps into editable grids. A rule changed over dinner is on the table the next evening.

AI let the games iterate at the pace of play. Seeing why a six-year-old gets stuck is still the craft.

Fig. 07 · Personal project

Music video2026

Clorinda

A six-minute music video that takes a canto of Tasso’s Jerusalem Delivered to young people who would never open the poem. Music, images and motion are all generated, and a written canon holds them together. Watch on YouTube

Role
Concept, direction, art direction, edit
Time
Two months, summer 2026
Format
16:9, 6:15, English and Italian subtitles
Tools
Suno, GPT Image, Seedream, Seedance, P-Video, Wan, lemon-gen

Why a classic

Most teenagers meet Tasso as homework, if at all. Yet the poem is a powerful story, written with a richness of image and rhythm that is rare today. A music video is a format they already watch, and the lyrics keep the poem’s own words, in Fairfax’s English translation from 1600.

Why it still speaks

The canto works on two levels. On the surface, a clash of civilisations over one city, between cultures more entangled than either side admits: Clorinda is the daughter of a Christian king, raised Muslim. Underneath, a knot no treaty can untie: personal stories. He loves the enemy he is killing.

Two times

In 1099 the night duel is silent: no one on screen speaks, no lips move. In the present, the same two play their story as a string quartet in the courtyard of the Tower of David, with Queen Melisende on cello and Tasso on viola. Every sung word comes from the present. They share a frame only in the last shot.

The music

The music was made by hand in Suno. There are no drums: hard, aggressive bowing carries the rhythm, so the musicians on screen are really playing what you hear. I rewrote the track three times as the edit showed what it needed: less battle, more room for the ending. The final version joins two separate generations, a relentless first half and a slow, sacred second half.

The split with AI

I owned the story, the rules, the casting, the selection and the edit. Agents kept the canon: every visual fact went into a written bible first and into prompts after. Reference images came before any shot, about ninety of them, then 27 batches, 227 stills and 141 clips. Every file can be traced to the characters, places and props it used, and every batch showed its cost before anything ran.

A model gives you a beautiful frame every time. A film needs the same face, the same helmet and the same dawn three hundred times.