Bachelor-/Masterthesis: State-Aware AI Agent for MaCoCo

The RWTH uses the datacentric information system MaCoCo, to keep track of employee data. Recently, an GPT based AI-Agent has been integrated into its web interface to respond to user queries. Utilizing a RAG approach, this Agent can access information regarding MaCoCo processes. However, the current state of MaCoCo is not accesable for the agent. Letting the Agent access state information such as, which employee is on holiday or which employee works on which project, would greatly enhance its usability.

So, in this master’s thesis, the AI agent should be meaningfully extended with new features enabling database interactions. The following are some suggestions for possible extensions:

  1. Let the Agent read from the database and incorporate read information in its answers.
  2. Let the agent execute database changing task when prompted
  3. Use an agent to detect user errors and notify them of better options
  4. Automate generating agents with some or all discussed features from generic databases
  5. Your Idea?

A suitable GPT instance will be provided. To implement the features, function calling (see for example OpenAI docs) might be used. It is the current standard method for enabling LLMs such as GPT-4 to communicate with external resources. For all new features, it’s essential to select an appropriate evaluation scheme and employ it to assess their performance against an established baseline implementation.

Your Profile:

  • Java
  • LLM prompting techniques
  • Monticore/Montigem from e.g. SLE lecture
  • Good written and oral communication skills in German or English

Contact

Interested in the topic? Find out more about our current research publications über unsere aktuelle Forschung.

For more information please send your application documents to Oliver Tautz, M.Sc..

Task Definition:

Prof. Dr. Bernhard Rumpe
Lehrstuhl Software Engineering
Ahornstr. 55
52074 Aachen