Bachelor's or master's thesis: AI agents for customer inquiries (m/f/d)

What it’s about

An email comes in: is it a complaint, a question about delivery times or a request for a new quote? People can tell in seconds. For a language model it is surprisingly tricky – especially when a thread is long, mixes several concerns, or mistakes are costly. That is where your thesis comes in: you investigate how an AI agent recognizes incoming inquiries, summarizes them, routes them to the right place, and at which points it deliberately hands over to a person.

vona is a web agency from Leipzig that works remote-first for clients across the German-speaking countries. AI integration, web and app development, design, and SEO & GEO all sit within one team. We use AI as a tool, not a gimmick – and we agree up front how success will be measured.

You can see what this looks like in practice in our AI customer advisor for online retail: it handles 68% of support inquiries autonomously, and first response time dropped from 18 hours to under 90 seconds. For another retailer, AI writes draft replies to returns and product questions, which the team reviews and approves. One possible research question: how can you draw the line between “the agent decides” and “a person decides” so that misrouted inquiries measurably go down without too many cases landing on the team’s desk? Your own ideas and focus areas are explicitly welcome.

Your tasks

• Work with us to shape a research question and study design that fit your university’s requirements.
• Build a prototype for pre-sorting, for example with LangGraph or n8n and models such as Claude, GPT or a locally hosted Llama via Ollama.
• Define a category scheme for inquiries and prepare an anonymized test dataset.
• Measure accuracy, misrouting and the share of cases handed to a person, and compare different approaches.
• Derive recommendations on when such an agent pays off for small and mid-sized companies and how to introduce it safely.

What you bring

• Studies in computer science, business informatics, data science, computational linguistics or a related field
• First experience with Python or TypeScript and an appetite for working with language models via an API
• You enjoy measuring things properly rather than just building a demo
• An interest in data protection and in how AI genuinely takes work off people’s plates
• Good German, as our clients and their inquiries come from German-speaking countries

Nice to have, not a must: experience with LangChain or LangGraph, a feel for evaluation methods, or a small AI project of your own.

What we offer

• A topic drawn from real client projects instead of a textbook exercise
• A dedicated contact at vona who supports you on the subject matter, plus regular check-ins; academic supervision stays with your university
• Access to current AI tools for your experiments, including Claude, ChatGPT, n8n and Ollama
• Flexible hours that fit around your semester, working hybrid – in Leipzig and from home
• If things go well, the possibility of continuing to work with us after your thesis, for example on follow-up projects; we can’t promise this in advance

How to apply

Just use the form at the bottom of this page. A few sentences about who you are, what you study and what draws you to AI agents are enough – plus a link to your CV, GitHub or LinkedIn. No formal cover letter needed. We usually reply within a week. Feel free to apply even if you don’t tick every box.