Use cues, state, invariants, and spaced rebuilding to turn coding problems into recognizable mental models.
Guides
Long-form pieces
14 guides, newest first. Opinionated notes on senior ML interviews, system design, and applied practice.
- How to learn coding interview problems without memorizing them
- Annotated senior-principal mock: ecosystem ranking
A synthetic ten-turn mock showing how an upper-IC candidate handles a watch-time win, creator damage, causal uncertainty, rollback, governance, and portfolio change.
- Annotated upper-IC mock: reasoning under a fixed budget
A synthetic ten-turn interview showing how fixed-capacity arithmetic, verification, serving, recovery, portfolio choices, and delegated leadership affect upper-IC calibration.
- Annotated principal architecture mock: enterprise agents
A synthetic interview transcript showing how technical framing, authority, failure semantics, portfolio judgment, delegation, and reversal move an answer from staff to senior principal.
- Frontier AI lab interviews in 2026: prepare for the format, not the logo
Confirm the round's environment, tools, expected artifact and scoring before choosing practice. Keep current instructions separate from dated reports.
- Proof of work: the bounded claim, artifacts and references
Make a personal contribution inspectable without claiming the team's work, exposing confidential evidence or scripting references.
- Designing a RAG system that actually works
RAG fails most often at retrieval, not generation. A practitioner's guide to the architecture, the failure modes, and what production teams actually do in 2026.
- LLM Evals: The hardest part of shipping LLMs, and why most teams get it wrong
Your model is only as good as your eval. Your eval is a product. Treat it like one. The patterns that separate teams that ship from teams that thrash.
- Senior through senior-principal ML scope
Calibrate upper-IC ML interviews by problem ownership, technical depth, portfolio judgment, delegated authority, durability, and evidence rather than inconsistent company levels.
- How to think about LLM inference cost
Count each billed model call once, including cache state, outputs, tools and retries; compare complete cost under quality and authority limits.
- System design case study: personalized search ranking
Design personalized search from retrieval through ranking, counterfactual evaluation, serving, experiments, feedback control, migration, and staff-level operating decisions.
- Five observable skills to practice across ML interview prompts
Use attempts and feedback to practice framing, relevant evidence, technical mechanisms, critique and collaboration without assuming a universal hiring rubric.
- Applied Scientist vs MLE vs Research Engineer: what these roles actually do
Companies use the same ML titles for different work. Compare the modeling, research, engineering, and product mix before choosing a target role.
- Lessons from Marin 8B: what an open pretraining log actually teaches you
Marin trained the first open-source 8B model to beat Llama 3.1 8B and published every mistake. The transferable lessons aren't about TPUs. They're about how to run pretraining like a science.