Industry

2025 to present

Multi-agent systems and GraphRAG for enterprise knowledge

Client projects at ponturo consulting where several AI agents work together and answer questions from company knowledge.

My role

As Senior AI Engineer, technical lead for small delivery teams and main technical contact for clients, from requirements and architecture to deployment.

What I built

  • Multi-agent systems with LangChain and Auto-Agentics on modular architectures.
  • RAG and GraphRAG solutions that combine knowledge graphs with long context handling.
  • Agentic workflows and full stack web applications on cloud native microservices, in the cloud and on premise.

Tools

  • LangChain
  • Auto-Agentics
  • RAG
  • GraphRAG
  • Knowledge graphs
  • Microservices

Industry

2025 to present

Document pipelines and AI agents connected to SAP

Automation of enterprise processes with AI, from reading documents to answering questions about business data.

My role

Developed as AI Engineer and continued as Senior AI Engineer at ponturo consulting.

What I built

  • Pipelines for document parsing, summarization, transcription and data extraction.
  • AI agents for statistical analysis, enterprise queries and reporting connected to SAP and SAP HANA.
  • Enterprise AI assistants on cloud native AI pipelines.

Tools

  • SAP
  • SAP HANA
  • AI agents
  • Document parsing
  • Transcription

Research

2023 to 2024

Cluster-Normalize-Activate modules for deeper GNNs

My Master's thesis research at the AI and Machine Learning Lab of TU Darmstadt, published at NeurIPS 2024.

My role

Thesis author and third of five authors of the paper. The central idea, method, experiments and results came out of my thesis.

What I built

  • A module that clusters node embeddings, normalizes each cluster and applies learnable rational activations per cluster.
  • MADC, a new metric for cluster dispersion.
  • Experiments in PyTorch on Cora, CiteSeer, PubMed, DBLP and ogbn-arxiv.

Tools

  • PyTorch
  • GNN
  • Rational activations
  • MADC
Research and publications

Research

2020 to 2023

Perception and obstacle avoidance for mobile robots

Research assistant work at the Intelligent Autonomous Systems Lab of TU Darmstadt.

My role

Research assistant. I also prepared and supervised robot learning exercises for students.

What I built

  • Deep learning models for visual perception and obstacle avoidance with encoder decoder architectures.
  • Model based reinforcement learning for adaptive robot behaviour, and models for object recognition and scene understanding.
  • Robot software in ROS, trained and tested in Gazebo and Isaac Gym, with grasp detection and motion planning in OMPL.

Tools

  • PyTorch
  • ROS
  • Gazebo
  • Isaac Gym
  • OMPL

Research

2021

An educational framework for robot learning

My Bachelor's thesis at the Intelligent Autonomous Systems Lab, supervised by Prof. Dr. Jan Peters and graded 1.0.

My role

Thesis author, from the design of the simulator to the exercises and documentation.

What I built

  • A Python robot simulator built with Pymunk and Pyglet, documented with Sphinx.
  • Exercises on kinematics, splines, PID control, Markov decision processes, ProMPs, REPS and ball catching.

Tools

  • Python
  • Pymunk
  • Pyglet
  • Sphinx

Industry

2020 to 2021

Web application at Eisenbahnbetriebsfeld Darmstadt

My Bachelor internship as a web software developer.

My role

Web software developer in an agile team.

What I built

  • A Go backend on BoltDB and a Vue.js frontend.

Tools

  • Go
  • BoltDB
  • Vue.js