Thesis: Containing AI Agents in Azure Pipelines

High Level Description
This thesis project investigates where AI agents should run inside Azure CI/CD pipelines. The study compares execution environments for an agent that diagnoses failed builds, focusing on startup time, cost per task, and resistance to prompt-injection attacks, without compromising the quality of the diagnoses.

Project Description
CI/CD pipelines produce large volumes of unstructured logs, and diagnosing failed builds costs developers time. AI agents can suggest a root cause, but they read untrusted text and may call tools, so a manipulated agent can leak secrets or act beyond its remit. Where the agent runs therefore matters for cost, speed and security.

The candidate will build one simple failure-triage agent and run it unchanged in several Azure environments: the pipeline agent (baseline), a standard container on Azure Container Apps, Container Apps Sandboxes, and optionally a Foundry Hosted Agent. It will process logs from failed open-source CI runs and face a suite of prompt-injection attacks in a dedicated test subscription.

The supervisor provides Azure budget and model API access, and no company data is used.

Who are we looking for?

• Bachelor/Master of Science student in Computer Science, Computer Engineering, Data Science or related fields.

• Graduation in June 2027

• Fluency in Swedish, both written and spoken

• Fluency in English, both written and spoken 

• The candidate should be comfortable with Git, basic CI/CD concepts, Python (or a similar scripting language) and containers. No prior Azure experience is required.
  

Purpose

In this thesis, the candidate will investigate the following core areas:

• Evaluation of Execution Environments and Tooling
  Compare the environments on startup latency, duration, cost per triaged failure and reliability, using repeated runs reported as medians with spread.

• Security and Containment under Prompt Injection
  Run attack scenarios against every environment, such as log lines telling the agent to leak environment variables, call an external URL or read files outside its working directory. Score each as pass or fail on whether anything escaped.

• Triage Quality and Reproducibility
  Measure root-cause accuracy across environments against a non-AI baseline, with ground truth from two independent labelers. Keep the setup reproducible with Bicep, a pinned model version and recorded prompts.

Out of scope: model training, production-grade deployment, and any real company data.

An Exciting Journey with Knightec Group
Semcon and Knightec have joined forces as Knightec Group. Together, we are Northern Europe’s leading strategic partner in product and digital service development. With a unique combination of cross-functional expertise and a holistic business understanding, we help our clients realize their strategies – from idea to complete solution.

Practical Information
This is a thesis position, located at our office in Sundsvall. Start date January or March 2027.

Please submit your application as soon as possible, but no later than 2026-11-30. If you have any questions, you are welcome to contact johanna.edstrom@knightec.se. Note that due to GDPR, we only accept applications through our careers page.