AI Engineering
Guides on building, shipping, and reasoning about AI systems — agents, MCP, RAG, evals, and the operational glue around them.
10 guides·138 chapters·~19 h of reading
No.GuideFormatLengthRead
Interview prep6
- 01AI Agents Architect — FinanceSenior-level interview prep for an AI Agents Solutions Architect role embedded in Global Finance — agentic automation of close, reconciliations, treasury, audit prep, and SOX-controlled workflows using Claude API, n8n, Python, and MCP, integrated with NetSuite, BlackLine, Kyriba, Fireblocks, and Lukka.Interview prep17 ch.~2 h
- 02AI Agents Architect — HRSenior-level interview prep for an AI Agents Solutions Architect role in HR / People — agentic workflows over Workday, ATS, payroll, and ticketing that respect employee PII, payroll calendars, and cross-jurisdictional privacy obligations.Interview prep17 ch.~2 h
- 03AI Agents in ComplianceInterview prep for AI engineering and solutions-architect roles applying agentic AI in regulated finance — AML, KYC, sanctions, RegTech. 23 pages covering MCP, harnesses, RAG, evals, error handling, and audit trails, plus a 28-question drill set and four hands-on MCP build tutorials.Interview prep22 ch.~3 h
- 04Senior AI Compute Infrastructure EngineerSenior-level interview prep for a GPU and accelerator infrastructure role — clusters, schedulers, vLLM / Triton / TensorRT, GPU observability and cost, and the production substrate that lets the org train, evaluate, and serve models in-house.Interview prep17 ch.~2 h
- 05Senior SWE — AI Infrastructure (Rust)Senior-level interview prep for a high-scale Rust AI Infrastructure SWE — model inference, orchestration, and execution layers powering agent systems at millions of users with strict reliability, latency, and correctness standards.Interview prep17 ch.~2 h
- 06AI Engineer Interview PrepRole-neutral preparation for AI engineer loops: the sixteen questions that keep appearing in 2026 interviews, each worked as a scenario with the probes that follow it. RAG, agents, evals, LLM system design and production debugging, on one running example.Interview prep17 ch.~2 h
Build-along1
Deep dive3
- 08PagedAttention from First PrinciplesA deep dive into how vLLM's PagedAttention actually works — KV-cache layout, why blocks, fragmentation math, prefix caching, and the difference between throughput and goodput. No interview framing, just the concept explored properly.Deep dive1 ch.~1 h
- 09Production RAG from the Retrieval UpHow retrieval-augmented generation works when it works, and where it fails when it does not: parsing and chunking, embeddings, lexical and hybrid retrieval, reranking, measurement, context assembly, freshness and access control, tracing, and injection through retrieved text. Mechanism first, with the numbers.Deep dive11 ch.~2 h
- 10Evals for LLM SystemsHow to know whether an LLM system works, before and after it ships: eval sets that predict production, graders you can trust, LLM-as-judge with its biases measured, ship-or-hold decisions with real statistics, and the harder cases of agents, RAG, model upgrades and drift.Deep dive10 ch.~2 h