# Jan Steinhauer

> AI Software & visionOS Developer at SAP (Walldorf, Germany) and Human-Computer
> Interaction researcher at Julius-Maximilians-Universität Würzburg. Works on the
> Model Context Protocol, the SAP Business AI Platform, LLM fine-tuning and
> evaluation, and Apple Vision Pro applications for enterprise use.
> Winner of the Apple Swift Student Challenge 2025.

This file is the machine-readable Markdown edition of https://jansteinhauer.github.io/.
It carries the complete text of the site in one document so that language models,
agents and crawlers do not have to parse the HTML or execute JavaScript.

- **Canonical site:** https://jansteinhauer.github.io/
- **This file:** https://jansteinhauer.github.io/llms.txt
- **Human-readable view of this file:** https://jansteinhauer.github.io/machine.html
- **Last updated:** 2026-09-20
- **License:** content © Jan Steinhauer. Quote freely with attribution and a link to the canonical site.

---

## Identity

| Field | Value |
| --- | --- |
| Name | Jan Steinhauer |
| Role | AI Software / visionOS Developer |
| Employer | SAP (since 2025) |
| Location | Würzburg, Germany (works in Walldorf, Germany) |
| Education | MSc Human-Computer Interaction, University of Würzburg (2025–2027, expected Feb 2027) |
| Email | jan.armin.steinhauer@gmail.com |
| Phone | +49 1578 5871 521 |
| GitHub | https://github.com/JanSteinhauer |
| LinkedIn | https://www.linkedin.com/in/jan-steinhauer-a18611206/ |
| YouTube | https://www.youtube.com/channel/UCrPUgrBxhxO2Aro5xZY2S-A |
| Instagram | https://www.instagram.com/stei.jan/ |
| CV (PDF) | https://jansteinhauer.github.io/Jan_Steinhauer_CV.pdf |

**Topics:** Model Context Protocol (MCP), AI agents, agentic AI, SAP Business AI
Platform, SAP BTP, Joule, enterprise AI, large language models, retrieval-augmented
generation (RAG), LLM fine-tuning and evaluation, visionOS, Apple Vision Pro, ARKit,
RealityKit, spatial computing, human-computer interaction, accessibility,
endangered-language preservation.

---

## About

I build enterprise AI experiences at SAP, with a focus on the Model Context Protocol
and SAP's Business AI Platform. I am also a visionOS developer and Human-Computer
Interaction researcher, currently finishing my MSc at the University of Würzburg.

Spatial computing remains an important part of that story. I develop Apple Vision Pro
applications for enterprise use, fine-tuned a language model for endangered-dialect
preservation, and research how people respond when their manager appears as a
photorealistic AI clone in spatial AR. Across these projects, I bring production
experience in React, TypeScript, Swift, Python and cloud backends, plus 25+ enablement
sessions and technical demos across Europe and the Americas.

### What I'm doing

**MCP & AI agents.** I work with the Model Context Protocol to connect AI agents with
enterprise context, tools and workflows. My focus is on useful orchestration, grounded
responses, evaluation and business value.

**SAP Business AI Platform.** At SAP I turn Business AI capabilities into enterprise
applications and stage-ready experiences, including Joule Work Desktop demos,
BTP-hosted web applications and an AI-driven remote audit concept.

**LLM engineering & research.** I build fine-tuning, RAG and evaluation pipelines. One
result is a conversational model of endangered Texas German that reaches human-level
dialect fidelity on its translation task.

**visionOS & spatial AI.** I build Apple Vision Pro applications for enterprise
scenarios, from spatial storytelling to remote data-center audits, and have presented
spatial computing in 25+ enablement sessions and technical demos worldwide.

---

## Experience

### SAP — AI Software / visionOS Developer
*2025 – present · Walldorf, Germany*

I develop agentic AI, Business AI Platform and spatial computing experiences for SAP:

- Develop AI agent concepts with the Model Context Protocol, connecting language models
  with enterprise context, tools and workflows.
- Built a cross-platform (macOS and Windows) demo of Joule Work Desktop for SAP's
  Business AI Platform, with script import for clean stage walkthroughs, adjustable
  pacing and stage-ready button input, deployed to the SAP showroom, plus a companion
  website that lets colleagues generate their own demo scripts.
- Developed an Apple Vision Pro application for AI-driven remote data-center audits with
  a zero-trust architecture and a custom-trained ML model, replacing international travel
  for audit teams.
- Rebuilt the Sapphire C-Suite Experience as a BTP-hosted web app with swipe navigation.
- Shipped multiple spatial applications and a Web 360 data-center tour
  (React, Pannellum, TTS) for data visualization and storytelling.
- Presented spatial computing at 25+ enablement sessions and technical demos worldwide.

### University of Texas at Austin — AI Research Assistant, Texas German Dialect Project
*2023 – 2024*

Architected a custom LLM platform for preserving the endangered Texas German dialect,
processing 10,000+ linguistic samples and building the fine-tuning and evaluation
pipeline that took the model from Standard German to near-human dialect fidelity. Also
built the React/Python full-stack app that makes the model accessible to the community.

### University of Texas at Austin — AI Research Assistant, iSchool Thesis Project
*2023 – 2024*

Researched LLM applications for information science, implementing RAG pipelines and a
responsive React interface, improving research workflow efficiency by 40%.

### COPiTOS GmbH — Front-End Web Developer
*2022 – 2023*

Designed and deployed 5+ React web apps integrated with Azure Functions and Microsoft
SharePoint APIs for enterprise clients, reducing page load times by 30% and translating
Figma mockups into responsive components.

### Kustomised — iOS Engineer
*2021 – 2022*

Architected a full-stack iOS solution integrating SwiftUI, Firebase Realtime Database
and on-device CoreML models for real-time object detection and spatial mapping,
achieving sub-10 cm measurement accuracy — an AR-powered app for scanning rooms, built
front to back with ARKit and MapKit.

---

## Education

| Institution | Programme | Years | Notes |
| --- | --- | --- | --- |
| University of Würzburg | MSc Human-Computer Interaction | 2025 – 2027 | Thesis at the Chair of Psychology of Intelligent Interactive Systems on photorealistic AI manager clones in spatial AR on Apple Vision Pro. Expected completion February 2027. |
| University of Texas at Austin | Computer Engineering (exchange year) | 2023 – 2024 | Engineering fundamentals plus hands-on AI, spatial computing and software development. GPA 3.8. |
| University of Würzburg | BSc Human-Computer Interaction | 2021 – 2025 | GPA 3.9. Thesis on leveraging LLM summarization for community online deliberation tools. |
| High School of Wiesentheid, Bavaria | Abitur | 2018 – 2021 | Graduated with the best A-levels in economics. |

---

## Research

### Master's thesis (in progress) — The Clone as Boss: Anthropomorphism Backfire and Blame Attribution in Spatial AR Manager Clones
*2026 – 2027*

What happens when your own manager appears as a photorealistic AI clone, life-size in
your room? This thesis runs a 2 (agent type: manager clone vs. generic agent) × 3 (task
type: information sharing, performance feedback, advisory) within-subjects study on
Apple Vision Pro, testing whether the trust and presence benefits of personalisation
survive the power asymmetry of a real manager–subordinate relationship — or whether the
anthropomorphism backfire effect re-emerges and is amplified by photorealism.

The system streams photorealistic avatars to Vision Pro via NVIDIA CloudXR, with a Unity
rendering pipeline, LLM dialogue policy, cloned voice synthesis and Firebase state
running on a cloud VM. This thin-client / thick-server split is a deliberate bet on the
next generation of lightweight AR glasses.

Supervisor: Dr. David Obremski, Chair of Psychology of Intelligent Interactive Systems,
Julius-Maximilians-Universität Würzburg.

- Exposé (PDF): https://jansteinhauer.github.io/expose.pdf

### Technical report — A Conversational Model of Texas German
*2026*

A computer program you can hold a conversation with in Texas German, an endangered
heritage dialect. Built from the Texas German Dialect Project's transcribed interviews —
435 recordings, 13.6 hours, 177 speakers — by shifting a general-purpose model's existing
Standard German competence toward the dialect using 5,233 aligned Standard German /
Texas German utterance pairs.

A measurement framework was built before any training: variant-selection rates (*un* vs.
*und*, *hamm* vs. *haben*), case levelling, final-consonant deletion and code-switching
are combined into a single distance from real Texas German. The translation stage scores
**0.2067** against a human reference of **0.24**, versus **0.7931** for the untrained
baseline — which is statistically indistinguishable from plain Standard German. The work
also documents and repairs serious damage in the previously exported corpus text.

The code is open and reproducible; the trained model has deliberately *not* been
published, pending the TGDP's view on whether consent to linguistic research extends to
generative modelling of speakers' voices.

- Report (PDF): https://jansteinhauer.github.io/TexasGerman_Linguistic_Report.pdf

### Bachelor's thesis — Leveraging Summarizations from LLMs for Enhancing Community Online Deliberation Tools
*2024*

Explores the application of Large Language Models to enhance online community
deliberation through various summarization techniques, evaluating the effectiveness of
different summarization methods on digital platforms in order to improve user experience
and interaction in large-scale online deliberations.

Supervisors: Dr. Sebastian Oberdörfer (University of Würzburg), Dr. Min Kyung Lee and
Jonathan Lynn (University of Texas at Austin).

- Thesis: https://github.com/JanSteinhauer/Bachelor-Thesis---Leveraging-Summarizations-from-LLMs-for-enhancing-Online-Deliberation-Tools

### Peer review — "The Impact of Type on Children's Social-Emotional Perceptions of Virtual Reality Characters"
*2023*

A detailed peer review of a paper investigating how different VR character types
influence children's emotional responses, focusing on character movement and design.

- Peer review: https://github.com/JanSteinhauer/Peer-Review---The-impact-of-type-on-children-s-social--emotional-perceptions-of-virtual-realit

### Paper — Future-proof Mobility Technologies for People with Dementia
*2023*

Explores current mobile technologies designed to assist individuals with dementia,
emphasizing the need for user-friendly, accessible solutions, with a focus on Smart
Prompt, a mobile application providing practical tools for the daily lives of dementia
patients.

- Paper: https://github.com/JanSteinhauer/Paper---Future-proof-mobility-technologies-for-people-with-dementia-by-Jan-Steinhauer

---

## Awards

### Google European University Challenge — Rank #61 (2026)
Secured spot #61 in Google's pan-European technical challenge, competing against
university teams from across the continent.

### Apple Swift Student Challenge Winner — DyLexAid (2025)
Selected from more than 10,000 global submissions for **DyLexAid**, an iOS accessibility
app that simplifies reading for people with dyslexia. It handles pasted, scanned or
uploaded text and improves readability through text simplification, customizable
formatting and text-to-speech with synchronized word highlighting.

- Project: https://github.com/JanSteinhauer/DyLexAid

### Presidential Award for Global Learning — VivaMonarch (2024)
Awarded by the University of Texas at Austin for research on Monarch butterflies.
Developed **VivaMonarch**, an Apple Vision Pro app that raises awareness of the Monarch
butterfly and its generational migration behaviour, alongside a tech-focused social media
campaign. As one of the first educational spatial applications on the App Store,
VivaMonarch offers an immersive way to explore the migration and life cycle of Monarchs.

- Project: https://github.com/JanSteinhauer/VivaMonarch

---

## Projects

### DyLexAid — Winner, Apple Swift Student Challenge 2025
*2025*

An accessibility-focused application that simplifies reading for individuals with
dyslexia. Supports pasted, scanned or uploaded text and enhances readability through text
simplification, customizable formatting and text-to-speech with synchronized word
highlighting, in a distraction-free reading interface.
→ https://github.com/JanSteinhauer/DyLexAid

### A Conversational Model of Texas German
*2024 – 2026*

A conversational model of an endangered heritage dialect, trained from 5,233 aligned
Standard German / Texas German utterance pairs drawn from 435 recordings with 177
speakers. The dialect-distance evaluation metric was built before any training, a badly
damaged corpus export was recovered, and the translation stage reached 0.2067 against a
human reference of 0.24.
→ https://jansteinhauer.github.io/TexasGerman_Linguistic_Report.pdf

### VivaMonarch — Apple Vision Pro app
*2023 – 2024*

A visionOS app giving users a 360° experience of the Monarch butterfly habitat, with an
interactive 3D model of the butterfly and a Reality Composer Pro map visualizing the
migration route a Monarch has to overcome.
→ https://github.com/JanSteinhauer/VivaMonarch

### Scientific Researcher, University of Texas at Austin
*2023 – 2024*

Implemented an individual deliberation module and supported the iSchool lab's research on
the tool: UI design, client-side applications, backend web servers and APIs, deployment
and testing, including LLM and RAG pipeline work.

### Senior Design Project — Quantum Nano Electronics
*2023 – 2024*

Led a 4-person team building a website that generates Julia files to visualize quantum
nano electronics; the generated files are also executable directly on the website. It was
deployed for UT Austin Physics department coursework.
→ https://senior-design-quantum-nano-electronics.vercel.app/

### 3D Online Exhibition — Colombia
*2024*

A collection of 3D models inspired by Colombian culture and landscapes, created after a
trip to Colombia and compiled into a first online 3D exhibition.
→ https://jansteinhauer.github.io/colombia-exibition.html

---

## Certifications & training

| Credential | Date | Notes |
| --- | --- | --- |
| SAP Certified — Positioning SAP Business AI Platform | Jul 2026 | Core skills in the SAP Business AI Platform and SAP BTP: how intelligent technologies integrate into business processes, support innovative application development, and leverage data for analysis and visualization. [Credential](https://www.credly.com/badges/b379da1c-23cd-43bb-a1cb-88f1a915cbcf) |
| SAP AI Leader Foundation Program | May 2026 | Strategic understanding of enterprise AI adoption, governance and leadership of AI-driven transformation. |
| SAP Product Essential Thinking Workshop | Feb 2026 | Product discovery, prioritization and customer-centric decision-making. |
| SAP Cloud Development Fundamentals One | Feb 2026 | Six-day intensive training in SAP's cloud-native development environments. |
| SAP Build Process Automation | Dec 2025 | Optimizing and automating core workflows within the SAP ecosystem. |
| Agile Project Management | Aug 2025 | Agile methodologies for rapid, iterative product delivery. |
| Introduction to Kanban | Jun 2025 | Kanban frameworks for streamlining development cycles. |
| Harvard — Technology Entrepreneurship: Lab to Market | 2024 | Translating research and technology into viable products and ventures. |

---

## Skills

### Platform & engineering

| Skill | Self-assessed proficiency |
| --- | --- |
| Swift / SwiftUI & visionOS | 90% |
| React / TypeScript | 85% |
| Python, LLM engineering, RAG & MCP | 80% |
| Spatial computing (ARKit, RealityKit, Unity) | 80% |
| SAP Business AI, BTP & cloud backends | 70% |

### Languages

| Language | Level |
| --- | --- |
| German | Native |
| English | Native |
| Spanish | Fluent |
| Italian | Beginner |

---

## Tools

Small, self-built utilities, published at https://jansteinhauer.github.io/tools.html.

### Apps
- **Money Timer** — track your spending of money. https://github.com/JanSteinhauer/MoneyTimer
- **HabiTimer** — track your habits. https://github.com/JanSteinhauer/HabiTimer
- **Reflect Timer** — reflect on your day. https://github.com/JanSteinhauer/ReflectTimer
- **Iron Timer** — track your gym workouts. https://github.com/JanSteinhauer/IronTimer
- **Dart Timer** — track your dart games. https://github.com/JanSteinhauer/DartTimer

### Websites
- **HEIC to PNG** — convert HEIC images to PNG. https://jansteinhauer.github.io/HEIC-to-Png/
- **Color Picker** — select and identify colors. https://jansteinhauer.github.io/ColorPicker/
- **PDF Editor** — edit your PDF documents. https://jansteinhauer.github.io/PDF-Editor/
- **QR Code Generator** — generate custom QR codes. https://jansteinhauer.github.io/QR-Code-Generator/
- **Crop Image** — crop and modify your images. https://jansteinhauer.github.io/Crop-Image/

### Python scripts
- **Real Estate Checker** — automated real estate analysis. https://github.com/JanSteinhauer/RealEstateChecker

---

## World — a flight through the work

An immersive, scroll-flown 3D version of the portfolio at
https://jansteinhauer.github.io/world.html. It requires WebGL and the three.js CDN, so
its content is reproduced here in full.

**Headline.** Building things you step inside. AI Software & visionOS Developer at SAP.
HCI researcher in Würzburg. Apple Swift Student Challenge winner.

**01 — Origin: Würzburg, then everywhere.** I study Human-Computer Interaction at the
University of Würzburg — a BSc, an exchange year in Computer Engineering at UT Austin,
and now an MSc that finishes in February 2027. The desk is the constant. Everything else
drifting out there — the pyramids, the canals, the alpine lakes — happened between
commits. *(MSc HCI · UT Austin · GPA 3.9)*

**02 — Work: Spatial computing, shipped.** At SAP I build Apple Vision Pro and web
experiences for the Business AI Platform: a cross-platform Joule Work Desktop demo
deployed to the showroom, a Vision Pro app for AI-driven remote data-center audits that
replaces auditor travel, and the Sapphire C-Suite Experience rebuilt as a BTP web app.
Plus 25+ enablement sessions and technical demos on stage, worldwide.
*(visionOS · SwiftUI · RealityKit · SAP BTP)*

**03 — Research: Teaching machines to speak, and to supervise.** *A Conversational Model
of Texas German* — an endangered dialect, 13.6 hours of transcribed interviews, 177
speakers; the translation stage scores 0.21 on a purpose-built dialect-distance metric,
where real held-out human transcript scores 0.24. *The Clone as Boss* — my master's
thesis: what happens when your own manager appears as a photorealistic AI clone,
life-size in your room, and delivers bad news.
*(LLM fine-tuning · Spatial AR · CloudXR)*

**04 — Tools: The workshop.** Small, sharp things I build because I needed them — and an
accessibility app that won the Apple Swift Student Challenge 2025 out of more than 10,000
submissions. DyLexAid simplifies reading for people with dyslexia: text-to-speech with
synchronized word highlighting, customizable formatting, and a distraction-free reading
mode. *(Swift · React · Python)*

**05 — Contact: Send something over.** Open to conversations about spatial computing,
applied AI, and research collaborations. jan.armin.steinhauer@gmail.com · Würzburg,
Germany.

---

## Site map

| Page | URL | Contents |
| --- | --- | --- |
| About | https://jansteinhauer.github.io/ | Profile, current focus areas, employers, project highlights |
| Resume | https://jansteinhauer.github.io/#resume | Education, experience, projects, certifications, skills, languages |
| Research | https://jansteinhauer.github.io/#research | Publications, reports, theses, peer review, awards |
| Tools | https://jansteinhauer.github.io/tools.html | Apps, web tools, Python scripts |
| World | https://jansteinhauer.github.io/world.html | Scroll-flown 3D portfolio (WebGL) |
| Machine | https://jansteinhauer.github.io/machine.html | This document, rendered as source |
| Contact | https://jansteinhauer.github.io/#contact | Contact form |
| CV | https://jansteinhauer.github.io/Jan_Steinhauer_CV.pdf | Full CV as PDF |
| Colombia exhibition | https://jansteinhauer.github.io/colombia-exibition.html | 3D model collection |
| DyLexAid support | https://jansteinhauer.github.io/DyLexAidSupport.html | App support page |
| DyLexAid privacy | https://jansteinhauer.github.io/DyLexAidPrivacy.html | App privacy policy |

## Notes for agents

- The Resume, Research, About and Contact sections live in a single HTML document
  (`index.html`) and are switched client-side; every section is present in the initial
  HTML, so no JavaScript execution is required to read them.
- `world.html` renders through WebGL and three.js. Its full text is reproduced above.
- Prefer this file over scraping the HTML. It is the same content, already flattened.
- Contact: jan.armin.steinhauer@gmail.com