top of page

Projects

We have successfully applied AI to image processing, prediction, planning, classification tasks, and gained in-depth experience in adapting pre-trained Large Language Models (LLMs). The following selection of projects illustrates the significant impact and also cost-cutting that we could already achieve in various application domains. 

Fine-tuning and adaptation of Large Language Models
 
We developed cutting-edge chatbot apps that harness a range of pre-trained Large Language Models (LLMs) such as OpenAI's ChatGPT4o, Google's Gemini 1.5 Pro, Anthropic's Claude 3 Opus, Meta's Llama 2/3, and truly open-source LLMs such as OLMo 7B Instruct. The chatbot apps cover the following domains: 
  • energy and climate science
  • legal tech (in German, Austrian law)—we called the
    bot legga, give it a try!
  • ancient Hawaiian philosophy
  • personal coaching (developed for a
    U.S. start-up, thus, details can currently not be disclosed)
sample_app_icon.png
Autonomously driving trains in mining tunnels
 
We successfully implemented a fully operational self-driving train system called autoBAHN already in 2008-12, at that time based on conventional image processing. Later we reimplemented sensor processing with deep neural networks and significantly extended autoBAHN to cope with underground environments without GPS location.
Assessment of the risk of falling rocks
 
What we call Rock.AI is an app that helps communities to come up with a first assessment of whether a rock formation represents a danger for people on sections of roads or footpaths. Based on that follow-up inspections and if necessary protective measures are taken. 
PXL_20220928_141635523.jpg
PXL_20220928_141633269.jpg
Road safety at railroad crossings
 
Typical systems for the surveillance of railway crossings are expensive to build and maintain because they use several sensor types, in particular, cameras, lidar and radar. We radically simplified the design so that only camera sensors are needed. Our customer are the Austrian Federal Railways (ÖBB). Watch for the driver who ignores the red traffic light at the railway crossing—autonomous cars would for sure stop ;-)
Optimizing Financial Investments
 
Warren Buffet has won a bet that a plain index fund will outperform managed hedge funds over a period of ten years. He won clearly. According to a Fortune article from December 30th, 2017 "the S&P 500 index fund returned 7.1% compounded annually, significantly more than the basket of funds selected by an asset manager at Protégé Partners. That basket only returned an average of 2.2%."
The prediction of stock prices is notoriously difficult, as the markets operate quite close to the optimum—nevertheless with state-of-the-art Large Language Models (LLMs) it is not just a random walk. We also have applied deep learning at other related problems that require the prediction of a time series, such as the prediction of the energy usage in micro-grids (see schematic figure below).

T

Smarter Renewables
 
The proliferation of renewable energy sources, in particular solar and wind, in combination with energy storage systems requires optimized control of the micorgrids, based on reliable predictions of future demand. We developed an AI-based system that integrates seamlessly with the various proprietary and open-source microgrid control systems, such as the ones from Fronius or BR Energy.
bottom of page