Daniel Buchberger | Applied AI · Knowledge Graphs

Location

Vienna · Steyr
Austria

Studies

BSc Business Informatics · WU Vienna

Focus

Data Science & Production Management

Role

Junior SAP Logistics Consultant · INFORMATICS

Currently — Bachelor Thesis

Researching multi-agent Knowledge Graph traversal as a grounding mechanism for LLM-based systems. Read more ↓

Research & Interests

What I explore.

Topics I actively work on, study or find worth understanding more deeply.

Applied AI Systems

Building practical AI pipelines and agent architectures for real-world automation tasks in enterprise and personal contexts.

LLM Agents & Orchestration

Multi-agent systems, tool use, memory management and reasoning patterns for reliable LLM-based automation.

Knowledge Graphs

Graph-based knowledge representation using RDF, OWL and SPARQL; building and querying structured knowledge bases with Neo4j.

GraphRAG & Enterprise Knowledge

Combining retrieval-augmented generation with knowledge graphs to ground LLMs in structured enterprise data.

Fullstack Development

End-to-end system design spanning backend logic, APIs, workflow automation (n8n) and self-hosted infrastructure.

Network Technology

Telecommunications, network-aware system design and embedded systems — grounded in an engineering education at HTBLA Steyr.

Enterprise Process Automation

Automating SAP-based logistics and accounting workflows using AI agents, local LLMs and structured integration layers.

Data Science & Machine Learning

Applied ML methods, NLP and statistical analysis with a focus on business informatics and production management contexts.

Projects

What I’ve built.

Technical and academic artefacts — each a response to a concrete question or problem.

Code editor on screen — Photo by Ilya Pavlov / Unsplash

INFORMATICS · 2025

AI Agent Invoice Automation

Prototype for an AI agent-based automated invoice processing pipeline with direct SAP integration. Demonstrated at the 20th INFORMATICS SAPshots event alongside manufacturing and healthcare use cases.

Question How can LLM agents support structured invoice processing in enterprise SAP environments with and without human intervention?

Methods LLM prompting, document parsing, workflow automation, on-premise model deployment, SAP API integration.

Outcome Built and validated a proof of concept for on-premise LLM and SAP integration as a viable enterprise architecture; presented publicly at SAPshots.

AI Agents SAP Python On-premise LLM

Interview Detection & Search Optimisation

Developed with three fellow students — a two-stage AI system that identifies interviews within the APA news database and optimises search relevance. Trained exclusively on secure, private infrastructure.

Question Can a two-stage model reliably detect interview formats and improve relevance ranking across a large news archive?

Methods Text classification, NLP-based feature engineering, relevance scoring, evaluation on private APA infrastructure.

Outcome Two-stage system improving both interview classification accuracy and search result relevance in the APA archive.

NLP LLM-Finetuning Machine Learning Python Secure Infrastructure

View on LinkedIn →

Minimal desk setup — Photo by XPS / Unsplash

Family Business · Ongoing

Paperless Digital Office

Built a fully paperless office using paperless-ngx, with automated invoice pre-accounting via local LLMs running on Ollama. Document classification and data extraction handled through n8n workflows.

Question Can local, self-hosted LLMs reliably extract and pre-account structured invoice data without any cloud dependency?

Methods LLM-based extraction via Ollama, n8n workflow orchestration, paperless-ngx document management, Docker-based self-hosting.

Outcome Less paper in daily operations; invoice pre-accounting automated via local LLMs, saving multiple hours per month.

Ollama n8n paperless-ngx Local LLM Docker

Research

Bachelor Thesis.

Graph-grounded question answering — at WU Vienna.

Title

Reducing Hallucinations in LLM-Based Question Answering through Hybrid Multi-Agent Knowledge Graph Traversal

In Progress · 2026

LLMs can sound confident and be completely wrong. This thesis asks whether grounding answers in a large-scale, human-curated knowledge graph — traversed in parallel by multiple agents — can change that. Evaluated in a domain where being wrong is not just a low score.

Research Focus

Hybrid graph traversal with parallel sub-agents
Hallucination reduction & answer faithfulness
Human-curated vs. LLM-extracted knowledge graphs

Technologies & Methods

Neo4j · Wikidata · SPARQL
LangChain · LlamaIndex
Ollama · vllm · open-weight LLMs
Python · MLflow

Education

Background.

2022 – today

Vienna University of Economics and Business (WU)

Bachelor of Science · Business Informatics

Specializations: Data Science & Production Management.

2016 – 2021

HTBLA Steyr

Engineering Diploma · Electrical Engineering & Communications Technology

Telecommunications engineering, PCB design and embedded systems.

Graduated with Distinction

Skills & Stack

Technical Profile.

Technologies and domains I work with across enterprise, AI and infrastructure contexts.

Enterprise & SAP

SAP MM SAP PP SAP S/4HANA ABAP

AI & LLM

LangChain LlamaIndex Ollama vllm GraphRAG RAG

Knowledge Graphs

Neo4j Cypher RDF / OWL SPARQL

Programming

Python SQL Java C

Data Science

pandas NumPy scikit-learn

Automation & Infrastructure

n8n Docker Linux paperless-ngx Home Assistant

Observability

Grafana Prometheus Loki Tempo Alloy

Networks & Hardware

Telecommunications PCB Design Embedded Systems

Contact & Links

Get in touch.

For academic exchange, technical discussions or project-related questions, feel free to reach out via LinkedIn, GitHub or Bluesky. I am also reachable by email.

LinkedIn ↗ GitHub ↗ Bluesky ↗