Free playbooks for engineers putting AI agents to work.
Each playbook takes one job, such as setting up Claude Code on a team repo or preparing for an AI engineer interview, and walks it through on real code.
Split one coding agent into a planner, an engineer and a reviewer, with scoped tools, a hand-off contract and caps, then trace the team against the agent it replaced.
LLM summarization strategies: pick the shape, then check it
Choose extractive, abstractive, hierarchical, map-reduce or refine by the job, keep every claim tied to the source, and judge summaries without a reference.
Filtering and structured LLM output: schema, repair, ground
Get JSON another system can trust: a short strict schema, Pydantic validators, a bounded repair, a grounding check, a filter on every reply and the rates you alert on.
Advanced query rewriting for RAG: fix the question first
Read the shape of the question, rewrite only the shapes that need it with HyDE, fan-out, decomposition or step-back, and prove each rewrite beats the raw question.
Systematic debugging: a forensic guide to bugs that come back
Reproduce on a clean checkout, bisect in time and space, instrument instead of guessing, test one hypothesis at a time, and write the post-mortem that kills the whole family of bugs.
AI engineer interview prep guide: the bar, the design round and a four-week plan
Calibrate to your level, answer the questions every loop asks, structure an LLM system design in five slots, tell stories that end in a metric and build a portfolio that gets callbacks.
Advanced document processing for RAG: layout, tables, code and images
Parse real PDFs with their columns intact, store tables as rows, chunk code by function, describe figures with a vision model, and keep every heading in its chunk.
Data preprocessing for RAG: clean, dedup and gate before you embed
Most RAG quality problems are ingest problems wearing a retrieval mask. Clean each document type, normalize per index, dedup at ingest and reject bad data before you pay to embed it.
Run the agent on events, give it typed tools and tiered memory, score it on five dimensions, and ship it behind flags, a sandbox, a kill switch and budgets.
Declare the budgets, wall off slow dependencies, fall over to a measured second provider, trace every turn, and change prompts and models without a rewrite.
Vector databases for RAG: pick, tune, filter and migrate
Pick a store on your hardest axis, tune HNSW by measuring recall, wire hybrid search, keep filters from eating results, size the memory and switch stores without downtime.
The RAG cheatsheet: chunk, retrieve, rerank, evaluate
The production RAG reference card: chunking by document shape, hybrid retrieval with RRF, when to rerank, an eval loop that blocks regressions, and the failures vector search cannot fix.
Normalize Unicode, split BM25 and vector cleaning, route languages, repair mojibake and redact PII once at ingest, so every retriever downstream sees honest text.
Production coding agents: a harness engineering guide
Build the harness around the model: a streaming loop, permission-checked tools, a shell with limits, context that survives long sessions, and evals that turn production ready into a number.
Limits, a sandbox, an allowlist and an audit log: the shell layer that lets a coding agent run commands without risking the host, the secrets or the network.
Stream every token and tool call over SSE, show each call as a card, keep the plan in view, and let people cancel, redirect and approve without losing the run.
