← Interview experiences
Rejected

AI Engineer at EY

Difficulty
Process took
1-2 Weeks
Rounds
3
Format
Hybrid
Applied via
LinkedIn

How it went

Before the Interview:

  • Take a deep breath and use prep time wisely - don't rush into the interview. Spend a few minutes gathering your thoughts and reviewing key points from your experience.

  • Practice communicating your experience - use AI tools to generate potential questions interviewers might ask about your projects and work. Practice answering them out loud until it feels natural.

  • **Get hands-on with GenAI fundamentals:**Learn how to make API calls to OpenAI and other LLM providers

  • Build basic agents and understand tool-calling patterns

  • Create simple implementations so you understand the mechanics, not just the theory During the Interview:

  • Pause before answering - it's okay to take a moment to structure your thoughts

  • Ask clarifying questions - if something is unclear, ask. It shows you think carefully rather than rushing to answer

  • Communication trumps technical perfection - clearly explaining your thought process is more valuable than having all the answers Key Mindset: Treat interviews as technical conversations, not interrogations. The goal is to demonstrate how you think and solve problems, not to prove you know everything. Confidence + clarity + curiosity will take you further than just technical knowledge alone.

What they would tell you

Before the Interview:

  • Take a deep breath and use prep time wisely - don't rush into the interview. Spend a few minutes gathering your thoughts and reviewing key points from your experience.

  • Practice communicating your experience - use AI tools to generate potential questions interviewers might ask about your projects and work. Practice answering them out loud until it feels natural.

  • **Get hands-on with GenAI fundamentals:**Learn how to make API calls to OpenAI and other LLM providers

  • Build basic agents and understand tool-calling patterns

  • Create simple implementations so you understand the mechanics, not just the theory During the Interview:

  • Pause before answering - it's okay to take a moment to structure your thoughts

  • Ask clarifying questions - if something is unclear, ask. It shows you think carefully rather than rushing to answer

  • Communication trumps technical perfection - clearly explaining your thought process is more valuable than having all the answers Key Mindset: Treat interviews as technical conversations, not interrogations. The goal is to demonstrate how you think and solve problems, not to prove you know everything. Confidence + clarity + curiosity will take you further than just technical knowledge alone.

How to prepare

Based on my experience across all three rounds, here's what I'd recommend: 1. Master Your Professional Story

  • Practice discussing your work experience naturally and confidently

  • Be ready to defend design decisions you made in past projects

  • Interviewers will go deep on "why" you chose specific approaches, not just "what" you built 2. Azure AI Services (Critical for Round 2)

  • I did a crash course on Azure AI offerings before the interview - this was a lifesaver in Round 2

  • They expect hands-on knowledge of cloud services, not just theoretical understanding

  • Know the trade-offs between different Azure/cloud options 3. Communication > Technical Knowledge

  • Asking clarifying questions is MORE important than immediately answering - I learned this the hard way in Round 3

  • When stuck or unclear, pause and ask for clarification instead of fumbling through

  • It shows maturity and clear thinking 4. RAG Deep-Dive (Most Important for Round 2) You need solid understanding of:

  • Vectorization: embeddings, model selection

  • Vector databases: which ones, when to use them, trade-offs

  • Chunking approaches: strategies and their impact on retrieval quality

  • Reranking techniques: improving relevance

  • Infrastructure considerations: deployment, scaling, monitoring 5. LLM Fundamentals

  • Understand basic parameters (temperature, top-k, top-p, etc.)

  • Know popular open-source models and their characteristics

  • Be ready for infrastructure-level questions about deployment and scaling 6. System Design Mindset

  • Think production-ready: error handling, monitoring, observability

  • Practice explaining bug scenarios and debugging strategies

  • Focus on real-world trade-offs, not just textbook answers Bottom Line: Understanding fundamentals and communicating clearly matters more than knowing everything. When in doubt, ask questions - it's a conversation, not an interrogation.

Their background

Software Engineer with expertise in AI/ML tooling and developer productivity. Built AnalytiQ (data quality & AutoML platform with GenAI), Prompt ACE (Chrome extension for prompt management), and sfinance (PyPI package for financial data extraction). Passionate about creating practical tools that streamline workflows, from data preprocessing to AI-powered enhancements

Round by round

  1. 1

    L160 min

    Part 1: Technical Discussion (30 mins) Started with a brief introduction, then dove into my professional experience. The interviewer was particularly interested in my GenAI and AI work since that was a key requirement in the JD. I discussed:

    • My work on the GA Call Transcripts Pipeline where I scaled NLP inference using Azure OpenAI Batch API to process 6-8K daily transcripts

    • The Fare Rules Extraction project where I deployed an open-source LLM with vLLM, reducing processing time by 95%

    • How I've integrated AI/ML across production systems at Scikiq The conversation was collaborative - they asked follow-up questions about architectural decisions, scale challenges, and how I chose between different AI approaches. Part 2: Live Coding (30 mins) Problem: Build a CLI chatbot with:

    • LLM-powered Q&A

    • Physics calculator (coefficient of restitution) with error handling

    • PDF report generator for research queries Approach: I was allowed to use Google and documentation. The key focus wasn't just writing code, the interviewer wanted to understand my thought process, how I break down problems, and how I communicate my approach. I completed the task. Overall Vibe: Very conversational and collaborative. They cared more about problem-solving approach and communication than perfect syntax.

  2. 2

    L260 min

    Part 1: Technical Deep-Dive (20-25 mins) Started with introductions, then moved into a detailed discussion about my professional experience. This round went much deeper than Round 1. The interviewer probed my understanding of cloud services - they weren't just looking for someone who had used Azure/AWS, but someone who truly understood cloud architecture, service selection, and trade-offs. They asked pointed questions about:

    • Which cloud services I've worked with and why I chose them

    • My hands-on experience with different components (storage, compute, orchestration)

    • Real implementation details from my projects at Scikiq This was clearly a key evaluation criteria - they needed someone with strong cloud fluency. Part 2: System Design - Building a RAG System (30-35 mins) Problem: Design a RAG (Retrieval-Augmented Generation) system at a high level. This was intense. The interviewer went very deep on:

    • Architecture decisions (vector databases, embedding models, retrieval strategies)

    • Scalability and performance considerations

    • Error handling and edge cases

    • Bug scenarios: They grilled me on potential failure points and how I'd debug and fix issues in production

    • What if retrieval returns irrelevant chunks?

    • How do you handle latency spikes?

    • What's your monitoring and observability strategy? The questions weren't surface-level "explain RAG" - they were testing whether I could make real engineering decisions and handle production challenges. Part 3: Candidate Questions (5-10 mins) I closed by asking intelligent, thoughtful questions about their tech stack, team structure, and current challenges. In my opinion, this saved the round - it showed genuine interest and shifted the dynamic from being grilled to having a two-way technical conversation. Overall Vibe: Challenging and rigorous. This was a proper technical deep-dive where they wanted to see depth of knowledge, not just breadth. Strong cloud experience and system design thinking were critical.

  3. 3

    L360 min

    What Happened: This was supposed to be a managerial round with one person, but 2 additional people joined unexpectedly, which completely threw me off. I already have a bit of a stutter, and the surprise change made me significantly more nervous. The Question: They asked about deploying an open-source LLM on GPU and how I managed scale. My Performance: Honestly, I fumbled this badly. There was miscommunication on my end - I didn't fully understand what aspect of "scale" they were focused on (horizontal scaling? request throughput? model optimization?). Instead of clarifying, I tried to answer and it came out unclear and disorganized. I could have handled this much better by:

    • Pausing to ask clarifying questions
    • Taking a breath to manage my nerves
    • Structuring my answer around my actual vLLM deployment experience Outcome: The round ended within 15 minutes, which was a clear signal it didn't go well. Honest Take: This was my weakest round. The unexpected panel format + my nervousness + miscommunication on the scale question created a perfect storm. I have the technical knowledge (I literally deployed LLMs with vLLM at Scikiq), but I failed to communicate it effectively under pressure. Lesson Learned: Always clarify ambiguous questions before diving into answers, especially when nervous. And mentally prepare for panel interviews even if they're scheduled as 1:1s.

What came up

MediumHybrid2 3 years

A candidate-reported account, lightly edited. Interview processes change by team and date.

Sources