Start here
Where to start, by what you're trying to do.
243 questions, guides, and concept notes can be a lot. Pick the path that matches why you're here. Each item links directly; come back to this page when you finish a path.
The three sections
-
Questions 76
The senior ML interview canon. Each one with what L4 / L5 / L6 answers actually sound like, the tells that get a strong-hire vote, and what gets you down-leveled.
Go here when: you have a specific question and want a level-calibrated answer.
-
Guides 10
Long-form opinion pieces on senior ML interviews, system design, and applied practice. Take a position and defend it.
Go here when: you want a senior-IC point of view on something contested (fine-tune vs RAG, evals, inference cost).
-
Concepts 157
Concise notes on attention, normalization, LLM internals, optimization, and the math underneath. Each follows the same template: definition, why it matters, mechanism, common confusions.
Go here when: you need to look up or refresh a specific concept fast.
Use what you read
Turn the core library into an executable plan
Choose a workload, attempt before reading, diagnose weak dimensions, and prove improvement with spaced retrieval and simulation.
1. Preparing for a senior ML interview loop
You have a loop in 2-8 weeks. You want to know what's actually tested at L5/L6 vs. what showed up on your old prep material.
- Start with the private readiness check, then choose an hours-based 2-, 4-, or 8-week plan
- Use the round-specific practice method on every question
- If you are interviewing at a frontier lab, verify the current format evidence and substitute only the matching work sample
- The 5 things AS interviews test for
- L5 vs L6 calibration: what's the difference
- Tell me about your most ambitious project
- How would you evaluate an LLM application
- Reduce LLM inference cost 10x
- Browse all 76 questions
2. Shipping LLM products or building agents
You're building a real LLM application or agent system. mlmentorship is interview-focused; my opinion writing on applied LLMs and agents lives on my personal site.
- Browse essays at hsaghir.com/blog (verification asymmetry, similarity-is-all-you-need, the-loop-is-the-product, under-specified coding agents).
- Skim Looplet, my iterator-first agent framework.
- For consulting on agents/evals/post-training: hsaghir.com/work-together.
3. PhD or postdoc transitioning to industry
You have research credentials. You want to know what industry interviews look like and how research projects translate to "ambitious project" stories.
4. Building deep technical foundations
You want the canonical ML/DL reference, organized so you can study for breadth and depth in parallel.
- Foundations: matrices as linear maps → SVD and PCA → eigenvalues
- Probability: MLE → KL divergence → Bayes rule
- Deep learning: backprop → attention → transformers
- LLM internals: FlashAttention → KV cache → speculative decoding
- Browse all 157 concepts by sub-category
5. Engineering leader / hiring manager
You're not preparing for an interview. You want to calibrate what good looks like, what to test for, or who to hire.
Lost? Press / to search. Or browse the three sections from the top nav: Questions, Guides, Concepts.