ALEX
SEIPEL
I work out where AI pays off.
Then I build it.
Ten years of engineering, from a two-person startup to a bank’s core platform. Hire me for a week of advice, or for the whole build.
10+ +
Years of experience
6
Industries
4
Years as co-owner & CTO
01
Services
Consulting and delivery01
AI consulting
I spend my working days inside a bank that is adopting AI tooling, and the honest answer there is never all of it. Which parts pay is what you hire me to find out.
- Assessment
- I sit with the people doing the work and watch where the hours go. Sometimes the finding is that no model belongs anywhere near the task, and you get that in writing too.
- Tool selection
- Buy one tool, wire several together, or build your own. I tell you which one fits, what a month of it costs to run, and what you commit to afterwards.
- Enablement
- A tool nobody opens is a subscription. I run the sessions and write the guidelines your team will follow, which is already my job at Commerzbank.
02
AI software
Off-the-shelf tools cover what every company has in common. Whatever is specific to your product is software, and that is the part I design and build.
- Better interface
- A chat box asks people to describe work they already know how to do. A screen built for the task is quicker, and it leaves fewer places for a wrong answer to hide.
- Model selection
- Between models the price differs by an order of magnitude, and so does the quality on your particular task. I test them against your data, and the expensive model goes in only where it earns the difference back.
- Your own documents
- Answering over your files is a retrieval problem before it is a model problem, and retrieval is where these systems usually fail. It is also where I started: my thesis was learning-to-rank for search.
- The wrong tool
- A model is often the wrong tool. If a database query or a better form solves it more cheaply, you hear that from me, even though it means I build less.
03
Software development
Most of the ten years came before this wave of models. I build web applications, APIs, online shops, and the internal tool everybody needs and nobody wants to own.
- Prototype first
- You get something you can click on early, while a change of direction is still cheap. Then we harden it into a product your team can maintain.
- Development with AI
- Agents write most of the code now, and I work with Claude Code and its peers daily. The craft moved into the specification and the review, and that part is still mine.
- Architecture
- AWS, Terraform, Docker, CI/CD. I once had two weeks to get a shop ready for a national television audience: auto-scaling EC2, Aurora, CloudFront and a WAF, all of it managed with Terraform.
- E-commerce
- I replaced Magento with headless Shopify, behind a Gatsby storefront I built, with subscriptions and Stripe on it. The marketing stack was mine as well, and so was the rest of that shop’s technology, for four and a half years.
- Accessibility
- Since June 2025 the European Accessibility Act has applied across the EU, and Germany implements it as the BFSG. A bank audits WCAG conformance rather than assuming it, and that is the standard I build to. Older interfaces I bring up to it.
02
Experience
2016 to nowI build software that has to hold up in production.
Ten years, from a two-person startup to a bank. Four and a half of them as co-owner and CTO at Brandl Nutrition, where I was also the only engineer. I owned the recipe engine, the shop, the fulfilment down to the printed label, and the AWS account under all of it.
Then a year at a Berlin consultancy, three clients in twelve months. I built a supply chain import tool, a design system for a testing and certification group, and a therapy platform for clinics. Today the work is a core banking frontend in Prague, the screen millions of customers reach after they log in. Before all that I trained neural networks at Immowelt, part of the Axel Springer group, and put them into production. The AI side is not new ground for me.
Off duty I lift weights and plan my meals in a spreadsheet, down to the gram. That habit is how I ended up at a nutrition startup in the first place, and it is how I build software: measure first, then decide.
- Location
- Prague, Czechia · CET
- Remote
- Worldwide
- Now
- Commerzbank AG · Engineer
- Before
- Co-owner & CTO · Brandl Nutrition
- Languages
- German · English
- alex@seipel.me
- Transaction Overview is the transaction list millions of customers reach after they log in. I am the only frontend engineer on it, and I moved its toolchain from React Scripts to Vite.
- For a new advisory platform I architected the React frontend from the first commit and built several modules on my own. It has AI features, and that is as much as I can say about it.
- Accessibility belongs to the definition of done. Every feature ships to WCAG conformance.
- I help drive the adoption of AI tooling in engineering and train colleagues on it. I onboard new developers, and I write the OpenAPI specifications with the backend teams.
- A sustainability startup: the frontend of an import tool for supplier and farm data, in React with AG Grid and GraphQL. My first direct client work, so I ran the requirements workshops before I built anything.
- A testing and certification group: a token-based design system, end to end. Figma tokens synced through Git, then web components in StencilJS, and wrapper packages that let the React, Angular and Vue teams pull one version.
- A medical technology company: a therapy platform for clinics worldwide. I had never written Angular, was productive within days, and mentored the junior developer on the team.
- I built the engine that turned a customer questionnaire into an individual supplement recipe. Behind it a Django backend produced the production sheet, the back label and the delivery note, then handed the parcel to DHL.
- A shot at “Die Höhle der Löwen” meant being ready for a national audience on two weeks’ notice. I moved the platform off one rented server onto auto-scaling AWS, managed with Terraform.
- I replaced Magento with headless Shopify, behind a Gatsby storefront I built. The marketing stack was mine too: analytics, A/B tests, newsletter automation, and the Python that read the numbers back out of them.
- As co-owner I had a say in the range too. The plain protein I pushed for became the best seller.
- Machine learning from the data to the deployment: price estimation, image classification, extraction from photographed energy certificates. TensorFlow, Keras, PyTorch, scikit-learn.
- Fraud prevention for one of Germany’s largest property portals, part of the Axel Springer group: a mobile TAN flow, and the identity-verification services behind it.
- “Markt und Preise” is a public tool where anyone builds their own charts over German rent and purchase prices. C# and .NET on ClickHouse.
- SQL Server, Vertica, Couchbase, MongoDB, ClickHouse. I wrote the real-time library that let a change in MongoDB start a pipeline downstream.
- My first commercial position: search engine optimisation for client sites, and the content that went with it.
- I built a search prototype on Apache Solr with learning-to-rank algorithms.
- The Twitter log data behind the experiments I analysed in R. That write-up became my thesis.
On the side
- impflotse.de ↗ (opens in a new tab)
- Patients book vaccination appointments at pharmacies. I am the only developer, I have worked AI-assisted since the first commit, and I built out the GDPR side too: legal logging, automated data disclosure.
- Affiliate SEO
- Earlier on, an affiliate site that ranked and earned. Running it is how I learned search engine optimisation.
From my employment references
“Mr Seipel’s performance earned our full recognition at all times and in every respect.”
“He found excellent solutions for all arising problems without exception.”
03
Stack
What I build withAI
- LLM APIs
- Claude
- GPT
- Agents
- MCP
- RAG
- Vector search
- Evaluation
- Open models
- TensorFlow
- scikit-learn
Languages
- TypeScript
- JavaScript
- Python
- Java
- C#
- SQL
Frontend
- React
- Next.js
- Angular
- Vue
- GraphQL
- Web components
Backend
- Node.js
- Django
- FastAPI
- .NET
- REST
- OpenAPI
- Stripe
- Serverless
Infrastructure
- AWS
- GCP
- Terraform
- Docker
- CI/CD
- PostgreSQL
- ClickHouse
- MongoDB
- Education
- B.A., University of Regensburg Media Computer Science and Information Science, 2013 to 2017. My thesis put learning-to-rank into a search prototype.
- Spoken languages
- German — first language English — level C1
- Clients
- Startups, Mittelstand, agencies, enterprise Solo, or inside a team you already have.
04
Approach
How I workNo slide deck.
No team of consultants.
One person, building.
- Scope
- We agree what is in and what is out before I start, and short cycles keep the cost visible and the exit cheap.
- Evidence
- I measure an AI claim before it gets a budget, mine included. Where the numbers do not hold up, you hear it from me first.
- Ownership
- The code, the infrastructure and the documentation are yours, and nothing runs on an account that only I can reach.
- Progress
- You see running software rather than a status report about running software.
- Leadership
- Where the technical lead is missing, I take it. I carried a startup’s whole technology for four and a half years as co-owner and CTO, with nobody else to hand it to.
Engagement
- Remote from Prague, CET
- German or English
- Day rate or fixed price
- From a two-day assessment
- To a build that runs for months
- Open for new projects
05
Contact
Prague · CET
Tell me about your project
alex@seipel.me