Last update: July 2026. All opinions are my own.

GenAI Engineering — Interview Prep · Part 3/15

The last five minutes of every interview: "do you have any questions for us?". Wrong answer: "no, you covered everything". Right answer: three or four specific questions per interviewer that show I've done the work, I understand what the job actually is, and I'm evaluating them as much as they're evaluating me.

This is my grouped list. I won't ask all of them — I'll pick the 3–4 most relevant to whoever's in the room.

Questions for the hiring manager

About the team and the day-to-day

  • What does a typical week look like for the AI engineer role you're hiring for? Notebook exploration, production PRs, client demos — what's the mix in a normal quarter?
  • How is the AI.DA CoE structured internally? Is it one team or several specialised pods (RAG, agents, MLOps, evaluation)?
  • Who would I be reporting to, and what does the reporting cadence look like — weekly 1:1s, sprint reviews, quarterly OKRs?
  • What's the size of the team right now, and what's the target size in 6 months?

About the projects

  • Which industries — banking, insurance, industrial — are driving the most GenAI work in the CoE right now?
  • Can you describe an active project (without breaking confidentiality) that this role would likely land on?
  • How many client engagements does a CoE engineer typically run in parallel? Is it deep on one or spread across two or three?
  • What's a project that didn't work out the way you expected, and what did the CoE learn from it?

About growth

  • What's the growth path for someone in this role — technical IC track, tech lead, principal architect?
  • How does GFT support certifications or conference attendance (AWS AI Practitioner, PyData, NeurIPS)?
  • Has anyone in the CoE published, spoken publicly, or open-sourced work? What's the appetite for that?

Questions for the technical interviewer

About the stack

  • What LLM providers are you deploying against most often — OpenAI, Anthropic, Azure OpenAI, self-hosted Llama / Mistral variants? Is the choice per-client or a CoE default?
  • Do you have a canonical RAG stack you reuse across engagements (vector DB, chunking library, reranking model), or is each project greenfield?
  • LangChain vs LlamaIndex vs writing it yourself — where does the CoE land, and why?
  • How do you handle model versioning and prompt versioning in production — is there a canonical way, or is each team building their own?

About LLMOps + reliability

  • What does observability for LLM-based systems look like at GFT — Langfuse, LangSmith, Datadog, a home-grown solution?
  • How do you set SLOs for a RAG system in a banking context, when "correct" is often subjective? What's the evaluation harness?
  • What's the on-call rotation like for a GenAI service in production? Who gets paged when the model starts hallucinating on a Friday afternoon?
  • How do you handle prompt injection and jailbreak defense today — allowlists, LLM firewalls, output filtering, a mix?

About cost + latency

  • What's the typical cost profile of a project — mostly per-token API costs, or infrastructure for self-hosted models?
  • Have you had to fall back from GPT-4/Claude to a smaller open model for cost reasons? What was that experience like?
  • What latency budgets do banking clients typically demand for user-facing agent workflows?

About the agent frontier

  • The JD mentions multi-agent patterns as a core skill area. Are you deploying multi-agent workflows in production today, or is that still mostly a demo / prototype phase?
  • What agent framework do you lean on — LangGraph, AutoGen, CrewAI, custom? And what's the reasoning?

Questions for HR / people-ops

Compensation + benefits

  • What's the compensation range for this role, and how is it structured (base + bonus, equity if any)?
  • Is there a formal review cycle for compensation adjustments, and how often?
  • What benefits does GFT offer around health, retirement pension, meal vouchers, transport? (Standard Spanish package questions — but ask specifically.)
  • Is there a training budget per engineer per year — for certifications, conferences, books, courses?

Location + hybrid

  • What's the hybrid / remote policy for this role? Days-in-office expectation, flexibility, off-site travel?
  • Which office would I primarily be based out of? Is there flexibility between the Madrid / Alicante / Sant Cugat locations?
  • How much client travel is realistic for this role — none, occasional, weekly?

Process

  • What are the next steps in the interview loop, and what's the realistic timeline to an offer?
  • Is there a take-home or case study round I should prepare for?
  • Who else would I meet before the offer — CoE lead, engineering director, teammates?

Questions for the CoE lead (if I meet them)

  • What are the top three engineering challenges the CoE is facing right now?
  • What's the shared playbook look like — is it a template repo, an internal wiki, an evaluation harness, all of the above?
  • How does the CoE decide what to standardise vs what to leave per-project?
  • How do knowledge-sharing rituals work — weekly show-and-tell, quarterly summits, internal blog?
  • What percentage of CoE time is client-billable vs internal R&D / tooling?

Red-flag questions (the ones that tell me to walk away)

I won't ask these directly, but I'll listen for the answers in what they do say:

  • If "how do you evaluate LLM output quality" gets a vague answer like "we look at it", that's a red flag — no evaluation discipline means no way to know when things are broken.
  • If "what's the on-call rotation" gets "we don't really have on-call for GenAI yet", that's not necessarily bad — it could mean the products aren't yet in production 24/7, which is fine, but I want to know.
  • If "how do you handle prompt injection" gets a wave-off, that means prompt injection defense is one of my first jobs, not a skill I'll learn from a senior. Not a deal breaker, but a signal.
  • If the answer to "how does the CoE share learnings" is anything less structured than a weekly ritual + a searchable knowledge base, the "CoE" is more branding than substance.
  • If salary transparency at the HR stage is bad, that predicts how the rest of the compensation conversation will go.

My top 5 (the ones I'll definitely ask)

Given time, I'll prioritise:

  1. What does a typical week look like for this role in a normal quarter? (calibrates the day-to-day)
  2. Which industries and what kind of client project would this role likely land on? (calibrates the actual work)
  3. What does observability + evaluation for LLM systems look like inside GFT today? (signals technical maturity)
  4. How do you handle prompt injection defense in production RAG systems for banking clients? (signals depth — I've studied this and want to know their answer)
  5. What's the growth path for a technical IC in the CoE, and what does the first 90 days look like? (calibrates my trajectory)

What's next in this series

The meta prep is done. From here, the series goes fully technical.