Skip to content
Work

What I have built — for clients and for myself

Seven pieces of work: two for clients, one co-founded application, and four products of my own, two of which I use in my own business. No success metrics — just what was built and what it does.


For clients

Training videos and commercials

Client project

Workplace Videos — training videos by QR code

Training videos in eleven languages, available right at the workplace: scan the QR code, pick a language, watch — no app, no password, and every playlist with its own key.

Technology used: Next.js · PostgreSQL (Neon) · Bunny Stream · DeepL · Vercel

Client project · Bahlsen PiCK UP!

Bahlsen PiCK UP! — commercials made entirely with AI

Three spots for TikTok, Instagram, and YouTube with a documentary, true-to-life feel — I handled all of the image and video production, with one main character who looks the same in every scene, and without a single day of filming.

Technology used: Images with Nano Banana Pro and GPT Image 2 · video with Kling and Seedance

Related service:AI Image Generation
Co-founded

Concept and development as a co-founder

CommTrain: home page with a practice conversation
Co-founded · concept and development

CommTrain — practising difficult conversations

Practise difficult conversations by speaking instead of ticking boxes — with an AI counterpart and a review afterwards.

What it does

  • Speak as in a real conversation. The AI plays a counterpart with a story of their own; wording, tone, and feelings shift with every sentence. You practise the conversation itself, not picking the right answer from a list.
  • Assessed on six dimensions (empathy, emotional validation, person-centredness, collaboration, structure, information) — for individual sections and for the conversation as a whole, so you can see what to work on next.
  • Methods drawn from research, including motivational interviewing — the review follows fixed criteria, not the whim of a language model.

Technology used: Next.js · Claude (conversation and review) · Groq Whisper (speech recognition) · ElevenLabs (voice) · Supabase · Sentry · Vercel

Own products

From AI agents to recipes

Pinspector: Pinterest search terms with search volume
Own products · used in my own business

Pinterest tools — Pinterest MATCHER and Pinspector

Every article gets its pin — created automatically, on the right board, on topics people actually search for.

What it does

  • Matching: each site's articles are checked against the existing pins, so you can see at a glance which articles don't have a pin yet.
  • Creating pins automatically: for articles without a pin, the image and text are generated; an AI picks the right board, and the tool publishes the pins on a staggered schedule.
  • Checking search terms: for every Pinterest search term, Pinspector shows the search volume, how it has changed over time, related terms, and the strongest pins on the topic.

Technology used: Next.js · Supabase · Claude (board selection, copy) · image generation (Runware, Ideogram) · Pinterest interface through a service of my own · Pinspector additionally DataForSEO and OpenAI

tailes.de on a phone: a story for your child
Own product · live

tailes.de — children's stories with an audiobook

A bedtime story of your own, with an audiobook — and your child in the leading role.

What it does

  • Briefly describe who the hero is and what's allowed to happen — and it becomes a story in chapters, suitable for children aged 3 to 12.
  • A painted cover illustration and a narrated audiobook version — for listening when there's no one to read aloud.
  • The first story is free and needs no account. No ads, and visitor statistics without cookies.

Technology used: Next.js · Supabase · language models from OpenAI and Anthropic · speech synthesis (OpenAI, Gemini) · Vercel

rAIzepte: photograph the fridge, the AI recognises the ingredients
Own product · live

rAIzepte — recipes from whatever is in the fridge

Photograph the fridge, get a recipe — the AI recognises the ingredients itself.

What it does

  • Three ways to a recipe: type in what you have, have ingredients suggested at random, or photograph the fridge — the AI recognises the ingredients itself.
  • A complete recipe with nutritional information and a photo, so you can see what to expect before you start.
  • Every recipe gets a page of its own and is filed by meal, diet, and cuisine — so you can find it again later.

Technology used: Next.js · PostgreSQL (Prisma) · OpenAI (text, image recognition, image generation) · Clerk · Vercel

Own system · in daily use

AI agents in round-the-clock operation — a server on the team

Agents that write, run, and monitor my applications — controlled through Telegram, including by voice message. A watchdog reports outages and restarts whatever has stopped, and each morning a single combined report lands on my phone.

Technology used: Claude Code on a Linux server of my own · Telegram · scheduled jobs · Supabase as memory · Cloudflare R2

More of my own projects

Smaller pieces from my own workshop

Italiano Tandem

Practise speaking Italian with an AI tandem partner who remembers earlier conversations — in the browser or through Telegram.

Market analysis

Condenses several market-breadth measures into one signal every morning: buy, hold, or stay out. A private tool, not investment advice.


Attila Arndt

Attila Arndt · Triple A Digital, Cologne

Which of these comes closest to what you have in mind?

Pick a time that suits you. In the intro call, we'll work out which process is worth tackling first — and whether I'm the right person for it.

Free intro call (opens in a new tab)

Or email me directly: hello@tripleadigital.de · I reply within 48 hours.