The Sorting Hat
For most of my PhD, the job search felt like a Sorting Hat ceremony from Harry Potter. Senior students would disappear for months, then reappear with their fates decided. Even as friends began graduating and landing roles, I knew little about what they were going through beyond the occasional proof of life.
When it was finally my turn, I found the process far more demanding than I had imagined — and I felt like I was learning the rules of the game while playing it.
In retrospect, a lot of my experiences were universal, and many things I learned now feel like common knowledge. This post is one data point for how the journey can look — hopefully making the job search a little less mysterious for someone not too long ago in my shoes.
Timeline
The source for this post is Alisa Liu's excellent blog post "Notes on the Industry Job Search". She applied for Research Scientist / Member of Technical Staff roles at the end of a 6-year NLP PhD at the University of Washington. Here's what her search looked like:
Not pictured: myriad informal networking conversations leading up to the search.
Company order matters less than you think. The common wisdom is to use a few companies for practice, then time processes so all offers arrive simultaneously for negotiation. In practice: practice interviews help but your stamina is finite — don't burn out before the companies you care about. External factors like headcount availability matter more than preparation. And deadlines have surprising flexibility — recruiters know you have other processes running. That said, some companies issue "exploding offers" with tight deadlines, so investigate norms before you start.
Getting the first interview. Try to do good work during your PhD, make friends, and collaborate a lot. Sometimes you need someone inside the company vouching for you. Be social at conferences, collaborate widely, attend networking events. During the job search, reach out to people you know — or don't know — and ask about opportunities. A big part of the search is reconnecting with people you may not have spoken to in years. This is normal, expected, and turns out to be a wonderful side effect of the process.
Interview Types
Technical skills and knowledge are evaluated much more than research experience, though the latter probably gets you the interview in the first place. Here are the categories Alisa encountered:
ML Coding
By far the most common. Implement a given architecture, a decoding strategy, a traditional ML algorithm, or something creative. Being fluent in PyTorch is a must. In rare cases you'll be asked to use only numpy — for instance, writing the backwards pass from scratch — but you won't be expected to know numpy syntax by heart.
General Coding
Basically LeetCode, sometimes with extra flavor. Build strong foundations here because the concepts often show up in ML coding interviews too.
Technical Discussion
No coding, but very technical. Two flavors: extended discussion around one topic (how you'd design experiments to answer a research question), or rapid-fire questions ("What is 5D parallelism? What's the difference between PPO and GRPO?"). The former tests how you think; the latter checks breadth of knowledge.
Research Discussion
The conversations we practiced most in our PhD. Start by telling the interviewer about a past project, then the discussion flows from there. They may ask about other papers on your CV. Tailor your research pitch to the role — interviewers are tired, so hitting the right keywords makes it easier for them to see your relevance.
Behavioral
Textbook behavioral interviews, plus occasional questions about AI safety or societal impacts. Enumerate memorable stories from your PhD and map them onto common behavioral questions so you can retrieve the right anecdotes instantly. Alisa failed her first behavioral interview because she assumed she was obviously well-"behaved" and came up blank on simple questions. Trust me — it is uniquely painful to reconstruct hazy memories while delivering them in an interview.
Math
Some companies have a math interview, ranging from fun logic puzzles to serious mathematical derivations with pen and paper. Brush up on probability, linear algebra, and calculus.
Job Talk
Shorter than an academic job talk, focused on a single paper or direction. Alisa's was all about tokenizers — she spent most of the time on a first-author work, then covered second-author and ongoing works briefly, as they tied together nicely.
Preparation
There is truly no better use of your time than studying for interviews. For Alisa, the experience was very much like being back in undergrad: taking notes, drawing diagrams, doing practice problems, and spending entire days in coffee shops making sure she understood fundamental ML concepts inside and out.
Technical interviews are hard, and the skills being tested require dedicated effort to develop outside of doing research. For her and for most people she talked to, the job search is a full-time job.
Key preparation advice:
- Stanford CS336 Homework 1 is crucial. Implementing and debugging a transformer comes up so often in interviews that it will pay off massively to turn it into muscle memory. It really isn't worth losing points on.
- Practice coding with AI assistance completely off. You will underestimate your reliance on AI tools otherwise. Interview settings don't allow Copilot.
- Each interview is unique. Build an intuitive understanding of scope from the description, the company's interests, recruiter hints, and company reputation. Think of each interview as a slightly different math or CS class you never attended, and now you have ~3 days to cram for the midterm.
- Sleep matters. Alisa did her first technical interview on 2 hours of sleep after cramming LLM inference all night — none of the last-minute knowledge came up, and she spent 10 minutes on an off-by-one error because her gears were barely turning.
- Take notes after each interview. They will be helpful for future studying and reflection.
Side benefits. Studying carried enormous side benefits. Having wider breadth of knowledge directly improved confidence as a researcher. She became more secure in conversations because she was less worried about gaps in her knowledge being exposed. She truly believes that if she had done some of this studying earlier in her PhD, it would have expanded the space of problems she could think about and have ideas in. Amazingly, studying also made her enormously more effective at her ongoing project — she was able to have technical ideas she never would have been able to access before.
Negotiation
The work is not nearly done after you receive your offers. Instead, there is a potentially extended period for learning more about your options and negotiating. It involves many conversations with potential future teammates and managers, lunch visits, and recruiter calls.
Negotiating is hard. Nothing in your PhD prepared you for this, and unlike interviews, this part cannot be conquered by studying. Compared to recruiters, you are outmatched in both knowledge of the market and the skill of negotiation. Everyone you talk to wants something different from you.
Practical negotiation tips:
- Lean on your friends for know-how of interacting with recruiters and for data points to calibrate your asks.
- Before every recruiter call, write down what you are willing and not willing to share, along with quotes you could recite verbatim.
- In the post-offer stage, anticipate questions they might ask and points they might make, and carefully construct responses that advocate for yourself while feeling natural to deliver.
- Be deliberate about every aspect of the process. It is time-consuming, but really worthwhile.
Conclusion
A huge part of the personal experience was managing all the emotions that come with being on the market. There is a lot of social perception to navigate: it is not a good feeling to compare yourself to your peers, everyone has opinions on where you should or shouldn't go, and people become unusually invested in how your life is going.
Navigating a huge decision space with incomplete information is stressful — small choices with no right or wrong answers (like who to contact when) have an outsized impact. Frankly, Alisa was stressed, miserable, and not functioning in other parts of her life for several months.
And here's the bittersweet part: the PhD is such a special time, where your only job is to have good ideas and execute them, to learn and grow as a researcher, without worrying about imminently securing a real job. While preparing for the future is important, cherish your PhD for the unique time that it is. These goals may be complementary — Alisa consistently found that she did her best work when she was having fun and chasing the questions her mind would not lay to rest.
Learning Resources
Here are the resources Alisa recommends, drawn directly from her experience:
- LeetCode 75 / Neetcode Blind 75 — General coding practice
- Stanford CS336: Language Modeling from Scratch — Breadth of LLM topics, Homework 1 is critical
- Self-Attention & Transformers — CS224N reading
- The Illustrated GPT-2 — Jay Alammar's visual guide
- Backpropagation — CS231N notes
- Introduction to Policy Gradient for LMs — Ivon's guide
- Lightweight Guide to Understanding GRPO and RL Principles
- How to Scale Your Model — JAX scaling guide
References
- "Notes on the Industry Job Search" — Alisa Liu, June 20, 2026. The primary source for this post.
- "AI PhD Job Hunt" — Nathan Lambert. Inspired Alisa's timeline visualization.
- Stanford CS336: Language Modeling from Scratch — Spring 2025 course materials.
- Alisa's Book of LLMs — Her continuously-updated LLM study notes.