Process: 1 - Practice your worst day every day; 2 - Never send a human to do machine's job; 3 - Exercise extreme ownership. Tech: 1 - Cell-based architecture; 2 - Chaos testing and fault injection.

How search evolved from SQL substring matching to production RAG: BM25 inverted indexes for lexical match, hybrid retrieval and reranking for semantic precision, and grounded LLM synthesis with citations.

A lab walkthrough of bootstrapping a small agentic team in Claude Code – one lead Claude that decomposes work, plus backend and frontend specialists running in parallel tmux panes, each in its own git worktree, talking peer-to-peer through a file-based mailbox. Includes the full bootstrap, a small POC task on this Flask blog, and the OOM lesson learned.

Top 5 ML models that earn their keep on a typical website – semantic search embeddings, gradient boosting for prediction, collaborative filtering for recs, density clustering for segmentation, and time-series forecasting for capacity – with notes on where to get them, how to install, and which knobs actually matter when tuning.

Similarities and differences. Touching hardware both APIs run on and strategies to load balance (and fallback).

MCP servers are the standard protocol to get insights across systems in AI-driven development. Which flavor works best (stdio, streamable http), how to call it (from the main thread or via agent), what are pros and cons of each approach.

Insights into the cost $$, practical implementation and when to pick which: blue/green at the infrastructure / cluster level (big platform moves, K8s upgrades, AZ shifts), canary at the service/app level for everyday deploys, and feature flags layered on top to decouple deploy from release.

When to use skill and when to go the agent route? These 2 functionalities are getting mixed up at times. Agents run in isolated context (they can't access your conversation history), while skills can, and that impacts output, isolation (or not) from main thread and tokens usage.