How I Built My Digital Twin — and What the Experiment Revealed About Me
A self-experiment: how I built a website with AI, analysed my business profile using anonymised data — and ended up asking unexpected questions about my own self-perception.
How I Built My Digital Twin
And what the experiment revealed about me
I'm about to walk into my next client call. The potential client asks me: "Could you send me something about your Adobe Workfront experience beforehand?" I send a link. He asks Frank DZ 0.1 directly. And within 30 seconds he gets a structured answer tailored precisely to his question — with concrete project references, my way of working, and an offer for an initial conversation.
That's still a future scenario.
Today I launched the digital twin in version 0.1.
But let's start from the beginning.
What I Actually Wanted: A Window Into What's Possible
The initial motivation wasn't primarily about my business website. I wanted to show what is genuinely possible today — with a structured AI approach, under real-world conditions, without shortcuts.
Specifically:
- A complete website with an AI backend — not assembled with a site builder, but built on a real tech stack without any coding skills on my part
- A personal profile generated from real data — but under strict privacy requirements
The second point mattered to me in particular. I used over 6,400 emails from ten years of business communication for the analysis. But not carelessly. All data was anonymised before analysis — client names, company names, personal details removed. Only then did the AI evaluate the patterns in my communication, my topics, and my working style.
That's not a nice-to-have detail. It's the prerequisite that makes an experiment like this credible — and at the same time an approach I'd recommend to any client working with sensitive data.
A Self-Experiment With an Unexpected Question
I'm fairly sure that digital profiles of me already exist somewhere. Anyone who is active online and writes emails leaves traces — even if you keep a low profile on social media, as I do. Those profiles come into being without my control, assembled from fragments others hold about me.
This self-experiment I wanted to run consciously and deliberately. Not to optimise or market myself — but to understand a tool that I and my clients will need in the future. And to answer a question that, honestly, I found genuinely curious:
How does an AI see me when it only knows my professional side?
I deliberately excluded the private dimension. What remains is the professional picture: my emails, my projects, my phrasing, my patterns of response. Do I recognise myself in that? Do I see things others perceive about me that I wouldn't have described that way myself?
That's a comparison I find interesting — between one's own self-image and what data says about you.
The Real Use Case: What Is Actually Possible Today?
I wanted to put down concrete proof — not a PowerPoint, not a theory. The question was: can you, today, without writing a single line of code yourself, build a complete professional website with an AI backend? And at the same time distil a personal profile from real data that makes an AI respond the way I do?
The answer is: yes. And that is the actual heart of this project.
At the same time, I needed a new website. My previous online presences all had the same problem: time-consuming to build, quick to go stale. And honestly — a classic "About Me" page with bullet points and a profile photo doesn't reflect how I actually work. I have conversations. I ask follow-up questions. I analyse before I respond. No static PDF can do that.
So I combined both: experiment + use case + new website — with an approach that genuinely interested me: Chat First instead of an About page.
Instead of a page that explains who I am, there is now a digital conversation partner that shows it.
The Tech Stack: Pragmatic, Not Flashy
I could have built this in a complex way. I chose not to — because the cleanest solution is rarely the most complicated one.
Obsidian as CMS — and as AI cockpit: I write in Obsidian anyway. My notes, project summaries, thoughts — everything lands there. Now the vault is the single source of truth for the website. New projects, updated copy, fresh insights — I edit in Obsidian, Claudian syncs, Vercel deploys. No separate CMS, no login screen, no extra tool.
What many people don't know: Obsidian isn't just my note-taking tool — it's also the cockpit for all AI orchestration. Via the Claudian plugin, Claude runs directly inside the vault — I can delegate tasks from within Obsidian without switching between tools. Claudian doesn't work alone: behind the scenes run specialised subagents — a web developer agent for code, a translator agent for EN translations, a knowledge-base builder for the profile. Each agent knows its domain; I coordinate the tasks. This isn't a science-fiction setup — it runs today, on my machine, inside my Obsidian vault.
Next.js + Vercel: Fast to go live, no DevOps overhead, excellent performance. Deployment runs fully automatically via GitHub. I genuinely looked for a European alternative — and found nothing comparable. As of today, Vercel is simply the leader in this segment. That may change — I'd welcome it.
Claude API (Haiku model): The brain behind Frank DZ 0.1. I deliberately chose the more affordable model — not because quality is irrelevant, but because a well-crafted system prompt outperforms an expensive model with poor instructions. The system prompt isn't a generic prompt. It's the distilled version of my working approach, my language, and my knowledge.
Bilingual (DE/EN): The website runs fully in German and English. Content written in Obsidian, translated by the translator agent, maintained separately.
How the Build Actually Went
This is the part that surprised me most — and at the same time demonstrates exactly what I've been explaining to my clients for years.
I didn't use Claudian as a developer who writes code. I used him as a structured executor who understands my requirements and implements them independently.
Concretely: in each session I described what I needed — functionally, not technically. "When someone asks about getting in touch, they should receive my real contact details, including a vCard download for smartphones." Claudian derived the technical implementation from that, adjusted the files, tested, and deployed.
That is system-context-first in practice: I define the what and the why. The how emerges from the context.
The result: a complete business website with chat interface, bilingual, with analytics, contact CTAs, vCard download, and an automated deployment pipeline — built iteratively, session by session.
What Frank DZ 0.1 Can Do Today
- Answer questions about my expertise — structured, concrete, with project references
- Present projects — 16 reference projects from 15 years, accessible via chips or free text
- Enable contact — with real data, CTAs for email, phone, LinkedIn, Telegram, Signal, and vCard download
- Communicate bilingually — in German and English, depending on the language of the page
- Be available around the clock — even when I'm with a client or in an initial conversation
What he cannot do: replace me. He's a door-opener, not a closing machine. Anyone who wants to go deeper talks to me.
An Honest Conclusion: Prototype, Not Product
I want to be transparent: Frank DZ 0.1 is a prototype. The profile isn't complete yet. If someone asks very deep, specific questions about Adobe Workfront — governance architecture in complex enterprise setups, for example — the twin will still hit limits today. The knowledge exists. The knowledge base is being continuously expanded.
But that's also the point. This project isn't a finished product I deliver once and then forget. It's a living system. And that is — if I'm honest — the most interesting aspect of it.
A classic website is done after launch. A digital twin gets better over time.
What I Take Away for My Clients
This self-experiment showed me what is realistically possible today — and where the limits lie. Both are valuable.
What works: Transferring structured knowledge into a conversational system. Anonymised data analysis under privacy conditions. AI as a multiplier for personal expertise, not a replacement.
What it requires: Good raw data, clear requirements, and a willingness to proceed iteratively. Anyone who thinks you hand a few texts to an AI and get a twin in return will be disappointed. Anyone who approaches it systematically gets a genuine tool.
I'll continue developing these experiences — not just for myself, but also for clients who want to unlock similar possibilities for themselves or their organisations.
If you want to know what's relevant for you — ask Frank DZ 0.1. Or ask me directly.
Frank Fischer is an interim manager, Adobe Workfront specialist and M&A consultant at Kreativraketen KG. To the chat →