Please note that we offer more topics than the ones listed below. These give you a rough overview of the areas in which we work. 

Come Together

The problem addressed in this thesis is that coordinating meetings among several people is still often handled through external SaaS tools, even when the process involves privacy-sensitive academic workflows such as thesis defenses. Existing scheduling tools solve the basic coordination problem, but they do not satisfy the RIS-Lab requirements of being fully self-hosted, private, and accessible to both humans and software agents.

The goal of this thesis is to extend the existing polls service at polls.ris.bht-berlin.de so that groups of people and agents can agree on a suitable meeting slot. A typical use case is that three people need to agree on a one-hour slot for a thesis defense. The system should support the full workflow from poll creation to slot selection and should expose an agent-friendly interface through an MCP server. A specific requirement is that the complete system remains self-hosted and private.

• Reference systems: https://whenavailable.com
• What will you learn: designing privacy-preserving self-hosted web applications, frontend and backend development, MCP server development, LLM-agent protocols, CI/CD, containerized deployment, cluster operation, monitoring, observability, and SRE practices for operating a small production service.

The Winner Takes It All

The problem addressed in this thesis is that online services need a controlled way to roll out new features and to evaluate whether one version of a feature performs better than another. Without feature flags and statistically sound multi-variate testing, new functionality is either released to all users at once or evaluated only informally, which makes product and engineering decisions less reliable.

The goal of this thesis is to build a self-hosted multi-variate testing and feature flagging system for services operated by the RIS-Lab. The system should allow developers to define feature flags, assign users or requests to variants, collect relevant metrics, and evaluate the results using appropriate statistical tests. The system should be tested using existing RIS-Lab services, for example polls.ris.bht-berlin.de

• Reference systems: github.com/growthbook/growthbook
• What will you learn: feature flagging architectures, controlled rollout strategies, A/B and multi-variate testing, experiment assignment and metric collection, statistical significance evaluation, frontend and backend development, integration with existing web services, system performance monitoring, and self-hosted DevOps operation.

The Book of Love

The problem addressed in this thesis is that module books contain important information about study programs, but this information is usually stored in semi-structured documents that are difficult to search, compare, and analyze systematically. This makes it hard to answer questions such as which modules overlap in content, which modules could be considered substitutes for one another, or how a study program changes over time.

At BHT, we are building a system for online storage and processing of electronic module books: modules.ris.bht-berlin.de. The goal of this thesis is to use the RIS-LLM platform at ris.bht-pages.de/llm/ to create a natural language interface for querying and analyzing the module book database. The system should allow users to ask questions such as: “Which modules have the largest content overlap and are therefore likely to be substitutes for one another?”

• Reference systems: none
• What will you learn: processing semi-structured documents, designing data models, building natural language interfaces over structured data, applying LLMs to information retrieval, frontend and backend development, database-backed application design, and containerized deployment.

Show Me the Way

The problem addressed in this thesis is that Key Performance Indicators (KPIs) are only useful if users can quickly understand where problems exist and what might be causing them. BHT uses KPIs to monitor the performance of its study programs, but raw KPI values alone are not sufficient for decision-making. Users need visualizations that make deviations, trends, and potential causes visible.

The goal of this thesis is to develop, based on the existing KPI system and real data at kpi-superset.project.ris.bht-berlin.de, a set of visualizations that allow users and agents to (1) quickly pinpoint issues and (2) perform an initial deep-dive analysis of the underlying problems. The thesis should focus not only on showing KPI values, but also on helping users understand which study programs, cohorts, or modules require attention.

• Reference systems: hochschuldaten.che.de/berlin/hochschulen/
• What will you learn: KPI analysis, dashboard and data visualization design, business intelligence workflows, relational database querying, drill-down analysis.

Wake Me Up Before You Go-Go

The problem addressed in this thesis is that professional software engineering teams rely heavily on DevOps and SRE tooling to ensure reliable operation of their services, but alerting systems often create either too little visibility or too much noise. If alerts are missing, incidents remain undetected. If alerts are too frequent or poorly prioritized, alert fatigue makes the system ineffective.

The goal of this thesis is to develop and deploy a RIS-cluster-wide alerting system that can proactively contact the responsible people when something breaks. The system should integrate with IP telephony and speech LLM technology so that alerts can be communicated through voice-based channels. A core aspect of the thesis is the design of a self-monitoring and self-healing alerting architecture with a strong focus on minimizing alert fatigue.

• Reference systems: github.com/target/goalert
• What will you learn: alerting system design, incident response workflows, monitoring and observability, alert prioritization and alert-fatigue reduction, backend development, integration with IP telephony and speech LLMs, CI/CD, and cluster-based deployment.

Prof. Dr. Zbigniew Jerzak
Zbigniew.Jerzak@bht-berlin.de

Impressum | Datenschutz | Barrierefreiheitserklärung