Master/Bachelor Thesis - GenAI for Network Device Configuration Templates

Join our Team

About this opportunity

Large networks contain thousands of routers, switches and firewalls from different vendors, each with its own configuration language. Configuration management systems hide those differences: an operator makes a high-level request, and the system produces and sends the exact commands required by the target device. Templates make this translation possible, but writing and troubleshooting them requires knowledge of the template format, device command syntax and the high-level model being translated. Generative AI could assist with both tasks. The key challenge is determining whether its output is actually correct. In this domain, candidate templates can be checked automatically for format, translation and compilation, execution against a simulated device, and whether the produced commands are correct.

What you will do

This thesis investigates whether generative AI can write correct device configuration templates and explain and repair templates that do not work. You will first build an automatic judge that evaluates candidates through increasingly demanding checks. On top of this foundation, you will develop and evaluate AI pipelines for template authoring and troubleshooting. The two parts share the judge and experimental setup, making the project suitable for two students who can focus on complementary areas.

You will have freedom to define areas of exploration within the project. These could include, but are not limited to:

•    Building a staged checker for template format, translation and compilation, execution against a simulated device, and command correctness.

•    Generating templates from task descriptions, target-device documentation and existing examples.

•    Comparing generated templates with expert-written templates and measuring how far candidates progress through the checks.

•    Investigating which input information matters by removing one source at a time.

•    Evaluating whether feedback-driven retries improve results and how many retries are worthwhile.

•    Diagnosing failing templates, proposing fixes, and testing them against faults with known corrections.

•    Categorising remaining failures and examining how well success in simulation predicts success on real devices.

The skills you bring

•    Knowledge in Computer Science, Computer Engineering, or a related field.
•    Experience with Scala or Java; Python or a similar scripting language is also relevant.
•    Interest in generative AI applications and APIs, such as Amazon Q, Gemini or ChatGPT.
•    Familiarity with XML and Docker/Kubernetes is beneficial.
•    Interest in computer networks, developer tooling and empirical evaluation.
•    Eagerness to learn and explore new concepts. No prior knowledge of network configuration protocols is required.

Extent: 2 students, 30 hp each
Location: Karlskrona
Preferred starting date: January 2027