Ericsson AI Engineer Interview Guide

I interviewed with Ericsson Hungary for a Senior AI Engineer position.

The process had three stages: an initial HR conversation, a technical interview, and a final managerial discussion.

The first round was mostly about my background, current role, previous projects, motivation, and why I was considering a move. It was relatively relaxed, although there were some technical follow-up questions about the AI systems I currently work on.

The second round was where things became much more technical. It lasted around an hour and focused on practical AI engineering, RAG, agentic workflows, LLM applications, and system design.

One question that stood out was: “How would you design a production-ready RAG system for a large enterprise with both unstructured documents and structured business data?” I had to walk through ingestion, chunking, embeddings, retrieval, reranking, generation, and evaluation. The interviewer then pushed further on what I would do when a query contains multiple intents and the retriever only finds information for one of them.

Another question was about agents: “When would you choose an agent over a deterministic workflow?” We discussed the trade-offs between flexibility and reliability and how I would make an agentic system observable and safe enough for production.

The interviewer also asked: “How would you make an LLM-based application reliable enough for production?” That led to a discussion about evaluation, guardrails, validation, retries, fallback strategies, monitoring, prompt/version management, and testing the full workflow rather than only the model itself.

There was also a discussion about RAG evaluation, including how to measure retrieval quality separately from generation quality and how to tell whether the retrieved context actually supports the final answer.

I felt very confident when the questions were connected to RAG, agents, evaluation, and LLM applications because these are areas I work with regularly. I could explain my decisions using examples from actual projects rather than giving textbook answers.

Where I started sweating was the broader system-design discussion. The interviewer kept asking “why?” and “what happens if this doesn't work?” rather than accepting the first architecture I proposed. Questions around scalability, failure handling, and production reliability required more careful thinking.

The final managerial discussion was less technical and focused more on team fit, communication, expectations, and how I approach engineering problems.

Overall, the interview felt challenging but fair. The biggest surprise for me was how much they cared about the engineering side of AI — not just whether I knew LLM or RAG concepts, but whether I could reason about reliability, scalability, evaluation, and the trade-offs involved in putting these systems into production.