Selection of Projects
My work
Over time my expertise evolved, and currently covers the following aspects:
-
Business Case Evaluation
-
Data: Acquisition, Preparation, Interpretation, Visualization
-
Machine Learning: Feature Engineering, Modeling, Training, Evaluation, Serving, Model updating, Monitoring
I work both as a coordinator of a team or create the solution alone depending on the project.
Here are a few samples from my latest work, especially around machine learning.
Smart appointments
(Text-nlp,
Prompt engineering, Fullstack)
Fullstack app integrating ChatGPT to WhatsApp and providing an AI-based smart scheduling bot. Programming instant messaging using Communication Platform as a service (CPaaS)
Techstack:
-
OpenAI ChatGPT API
-
Twilio API
-
Nylas Booking
-
PostgreSQL
-
REST API
Microchip Detection
(text-NLP,
Transformers-Deeplearning)
In cooperation with bee produced and the event series "Industry meets makers" I created a case study on how to improve the life cycle of electronic goods. After testing both public and private Large Language Models (LLMs), an individual solution was created making use of:
-
web scraping
-
natural language processing
-
setting up a model based on Bidirectional Encoder Representations from Transformers (BERT)
-
identifying microchips using Named Entity Recognition (NER)
Energy consumption
(Tabular data,
Outlier/Anomaly detection)
This project was done in the frame of my Master Thesis "Domain Analysis of machine learning operations: A model architecture".
An emphasis was put on the machine learning operations of an end-to-end outlier detection application. A classification model from energy data with a focus on renewable energy consumption was developed and made available as a web app. Throughout the development, best practices from the latest research were adopted.
Implementation details as open source available.
person localization in wilderness search and rescue
(Image data,
Image Analysis/Autoencoder-deep learning)
This is part of the exercise class "UE Computer Vision, Oliver Bimber / Indrajit Kurmi" at the JKU Austria.
The institute for computer vision has a specific research project for Search and rescue with airborne optical sectioning.
In this lab project, I implemented an unsupervised person localization algorithm.
Challenges within the implementation:
-
Data preparation of colored images from drones,
-
Color channel extraction and modeling of outliers
-
Autoencoder approach of modeling outliers
Implementation details as open source available.
Explainability of BERT as Visual Analytics
(Text-nlp,
Transformer-deep learning)
This was an explainable AI project calculating Shapley Values using the SHAP library. Shapley values in explainable AI allocate credit to each feature in a machine learning model, revealing their individual contributions to predictions, enhancing transparency, and enabling a deeper understanding of how the model makes decisions. It was used on the BERT Model, which is a language model pre-trained on a large corpus of English language data.
Classification of Explainable Artificial Intelligence according to Hohman et al.
Implementation details as open source available.
Protein Folding with alphafold
(Protein Data,
Attention network - deep learning)
This is part of the exercise class "KV Structural Bioinformatics, Alois Regl / Sepp Hochreiter" at the Johannes Kepler University in Linz, Austria.
In this project I focused on globin evolution, comparing sequences of myoglobin of various animals.
-
Visualizing the protein structures with Molviz.
-
Folding proteins with AlphaFold2 and Alphafold2-multimer.
-
Changing the Structure/Model/Chain/Residue/Atom (SMCRA) structure and predicting the protein again.
-
Using the Local Distance Difference Test (IDDT) for evaluation
Challenges within the implementation:
-
Protein visualization
-
Protein structure prediction
-
Manipulating Aminoacids