~/ashwani-bhat / interview-blogs

📚 Interview Blogs_

Field manuals for senior ML interviews — long-form, self-contained study guides with graded question banks, levelling rubrics, and one case study each, worked end to end with numbers. Written to be argued with, not admired.

agents · tool use · post-training

Agentic Harness — Senior ML Interview Field Manual

Everything outside the model weights that turns next-token prediction into an agent: the control loop, context engineering, tools, verification, evals, RL post-training, production economics, and security — ending in the Atlas on-call remediation case study.

95 questions 9 architecture layers 14 failure modes 1 worked case study read →
retrieval · evaluation · production

RAG — Senior ML Interview Field Manual

Retrieval-Augmented Generation interrogated properly: the pipeline stage by stage, advanced architectures, evaluation, production engineering, and a failure taxonomy — ending in the PolicyCopilot 40k-document case study.

74 questions 10 pipeline stages 12 failure modes 1 worked case study read →
attention · transformers · first principles

Attention — First Principles & Interview Field Manual

Attention built up from a weighted average: every shape derived, the maths interviewers actually push on, heads and masks and the Transformer block, positional encodings, cost trade-offs, plus interactive labs and an implementation you can audit.

120 questions 16 chapters 2 interactive labs Worked math + Python read →
retrieval · first principles · case study

RAG — First Principles & Interview Field Manual

Retrieval-augmented generation from the seven prerequisites up: the pipeline stage by stage, five calculations done by hand, advanced architectures, evaluation, production engineering, and a failure taxonomy — with similarity and memory labs plus a worked case study.

90 questions 16 chapters 2 interactive labs 1 worked case study read →