Over the past year, I’ve noticed a growing number of my coworkers making the leap into large model (LLM) and Machine Learning Engineering (MLE) roles. So I decided to take the long-term consulting program and leap out of my comfort zone one day.

🎓 Elite College Long-Term Career Consulting Program

After five years as a backend engineer at Meta, I decided to shift my career toward AI and recently accepted an offer from OpenAI. This is a short reflection on how I made that transition, the interview process, and what I learned along the way.

📍 Personal Background

• Undergrad: One of the top universities in Hong Kong

• Grad School: UCSD MS in CS

• Experience: 5 years at Meta, backend-focused with occasional MLE collaboration

• Goal: Transition into LLM/AI-focused engineering, with OpenAI as the target

• Joined: A highly selective career program that provides 1-on-1 guidance, insider-level support, and personalized training

🔧 Preparation Strategy

1. Deep Familiarity with My Own Projects
I made sure I could talk about every technical detail of my resume projects — especially those involving ML. I used the STAR method, and prepped one “anchor project” inside-out: data preprocessing, feature selection, model architecture, evaluation, and tradeoffs.

2. ML Systems Design Fundamentals
I revisited everything from the ML Design course I had taken — including architecture design, training pipelines, and serving models in production. I paid special attention to:

• Transformers, BERT, and GPT

• Multi-head attention (MHA), and common optimizations like KV-cache, GQA, MQA

• Normalization techniques (batch norm, layer norm, etc.)

3. LeetCode + Conceptual Mastery
I didn’t just solve problems — I focused on understanding variations, follow-ups, and how interviewers think when choosing questions.

4. Targeted Company Research
I studied OpenAI’s recent projects and publications, so I could have deeper conversations during the interviews.

🧠 Technical Interviews: What Came Up

Deep Learning Fundamentals:

• Gradient descent variants (SGD, Adam, etc.)

• Hands-on knowledge of BERT and GPT architectures

• Self-attention mechanism: concepts, code, and optimization strategies

• Data preprocessing: BPE, tokenization, masking, and dataset balancing

Real Interview Questions I Was Asked:

• Explain multi-head attention and when it becomes a bottleneck

• What’s the difference between BERT Base and BERT Large?

• How do you optimize GPT for long-context generation?

• What’s RoBERTa, and how does it improve on BERT?

• How would you design a training pipeline for fine-tuning LLMs?

System Design for ML:

• How to design scalable model training infrastructure

• Efficient data labeling and experimentation strategy

• Fault-tolerant architecture for large model inference

🤝 Behavioral Interviews

This round focused more on communication, leadership, and resilience:

• How do you collaborate with cross-functional teams under deadline pressure?

• Describe a technical decision you disagreed with — what did you do?

• Talk about a failed experiment — what did you learn?

One key takeaway: technical depth will get you into the room, but clarity, humility, and structured thinking get you through the behavioral rounds.

🧾 Offer Details & Reflection

• Level: L5

• Compensation: ~$950K/year

• Workload: Intense — more demanding than Meta, especially Oncall rotations

• Takeaway: Totally worth it — the learning curve is steep but exciting

I also received a lot of help during negotiation, including guidance on when to push and how to frame competing offers. It made a real difference in my final compensation.

💬 Final Advice

If you're a backend or data engineer considering a move into AI, this is the time to start. You already have the systems thinking and production mindset. Pair that with ML depth and you’ll be uniquely positioned.

OpenAI interviews were tough, but with the right guidance and preparation, they’re absolutely beatable.

A special thank-you to Elite College for the structured coaching, insider insight, and endless support throughout my transition. From mock interviews to system design drills and project polishing, I wouldn’t have made it to OpenAI without your guidance.