Josef Richter

Product Designer & Engineer

agentic design systems systems-level prototyping 15+ years designing products 15+ years building full-stack software

Enaia CRM

  • From MVP to mature product used by CBRE (world’s #1) and JLL (world’s #2)
  • Working daily in a trio with the CEO and the CTO
  • Complex broker workflows with unwritten rules and social dynamics turned into simple software
  • Fully agentic design system pipeline
  • Working prototypes in code, like the board below – try dragging the cards
Details
“I wish we’d had you when we started this.”
David Young
CEO, Enaia

T-Mobile Smart Home

  • Napkin sketch to consumer app rolled out in 8 countries
  • From 0 to 150 person team
  • New product concept and strategy, sold to management through working prototypes
  • Design system deployed across 6 scrum teams
  • Red Dot Award and iF Design Award winners
  • Apple adopted the same concept after us
Details
“It’s rare that you come across exceptional talent like Josef. He showed outstanding product design skills and a deep understanding of product strategy.”
Timon Reinhard
Lead Product Owner
T-Mobile Smart Home
T-Mobile Smart Home app screen
Red Dot Award winner 2022 — MagentaZuhause app iF Design Award — MagentaZuhause app

Medable Clinical Trials

  • Led the redesign of the participant app, used globally in 110 languages
  • Moved 20 to 30 year old scientifically validated forms to mobile without biasing trial data
  • Worked directly with scientists on every design decision
  • Rolled out new mobile & tablet design system across platforms
Details
“Josef is a bright holistic design thinker. He is able to break down the brief or problem into smaller clear parts, and provides clear paths to take.”
Sina Mossayeb
Chief Design Officer, Medable
Medable — Study Participant App Concept: a row of mobile screens from the redesigned clinical-trial app

Stock Sorted Shopify App

  • My own product, taken from 0 to 1: from discovering the problem in merchant forums to a live app on the Shopify App Store
  • Real paying customers, including a world-famous football club
  • Real-time shared inventory across stores, a concurrency problem Shopify can’t handle natively
  • Everything is mine: research, positioning, design, Elixir engineering, App Store listing, analytics and support
  • Not much of a visual design: the UI style is given by Shopify
Details
Stock Sorted — shared stock dashboard inside the Shopify admin

Advanced prototyping

Relativity: handling the model directly

I started with no actuarial background and built a working insurance-pricing model to explore a product idea: what happens to the interface when a complex calculation becomes effectively instantaneous?

  • Rapidly learned enough about insurance pricing, exposure, Poisson models, confounding and splines to model the problem credibly and explain it in plain English.
  • Went deep to identify what could be done differently
    • Found a focused technical change with huge UX impact:
      • A row-by-row refit would take an estimated 100–500 ms
      • Grouping equivalent policies and reusing the previous fit brings it down to about 3 ms
      • Crossed the threshold for smooth recalculation while dragging 💪 (16.7 ms per frame at 60 FPS)
  • Verified both the mathematics and the product experience through statistical tests, production-browser benchmarks, keyboard interaction, responsive design, and carefully designed light and dark themes.
Open the interactive prototype

FinViz

  • Simple but functional apps focused on the one feature in question; far cheaper to hit dead ends here than in development
  • Below: an AI copilot prototype I designed and built myself in a few days; a few years ago this would have taken a team a month
  • More on the subpage: a voice-controlled smart home and a 2017 conversational commerce app, both long before the LLM era
More prototypes

Agentic design system

I built a design system that an AI agent builds from – and then built the instrument to find out whether it actually works. Three screen-building tasks, given to Claude Sonnet 5 five times each, in three environments: with the design system, with its components but no written guidance, and with nothing. 45 runs, $41. Every number below is computed from an append-only hash-chained journal, not typed into a slide.

  • 49% fewer design decisions per element with the system than without – 1.44 against 2.82, ranging 13–77% across tasks
  • Half the styling vocabulary: 19 distinct CSS classes per screen against 40
  • Five runs of one identical prompt used 27 distinct styling decisions with the system, 114 without. The output without a system isn’t ugly – it’s a different product every time you ask
  • Zero governed components re-implemented by hand across fifteen runs. Without the system, thirty-three
  • And the result that went against me: my written guidance for agents measured as doing nothing – until a grader that reads composition found where its effect had been all along
How it was measured

Code

  • I don’t hand off static files. I ship working software.
  • I write code so I understand what’s easy, what’s hard, and what developers need from me; mostly Elixir, plus Rust, Swift and TypeScript/React
  • I build my own products alongside freelance work, often wrestling with high-concurrency problems (too many people → server crash)
  • Sorted – event registrations that don’t crash under heavy load. Fully agentic event management.
  • nutritious.love – AI calorie tracker: scan food, get nutrition-change recommendations, and more.
  • HabitPass – iPhone app that builds a language-learning habit by locking out social media.
  • Tideman – a tool that helps students test their implementation of the “final boss” algorithm from Harvard’s CS50.
  • francis_template – open-source Elixir package for file-based HTML templating in the Francis web framework (this very site runs on it).
  • francis_sqlite – open-source SQLite integration for the Francis Elixir web framework.

More of my work is on GitHub.