Search for a command to run...
A public embedding pipeline, two injected failures, and the job-level runtime a thread pool does not provide. You have 50,000 document chunks to embed. Eight workers are sending batches to an API. The
The main lesson is simple: retrieval for agents is not just “search, then answer.” It is a control system. The tools you expose, the names you give them, the shapes of their outputs, and the order the
Contract tests for the gaps between your linter and production We ran ruff check --fix on a Python service. 71 unused imports removed. Tests passed. CI green. We nearly shipped a change that would ha
Part 2: Building a Semantic Highlighter from Scratch Chapter 1: The Foundation - Understanding Our Tools Let's build our semantic highlighter piece by piece. I'll be your guide, and we'll construct each component while understanding why it exists. Fi...
Part 1: Why Highlighting Matters and How Machines Learn Relevance The Problem: Finding Needles in Haystacks Imagine you're searching for "how to prevent memory leaks in Python" and you get back a 5,000-word article about Python programming. Somewhere...
Introduction In Python development, we frequently work with structured data: configuration settings, API responses, form submissions, event logs, and more. Most developers reach for one of two common approaches: # Dictionary approach user_data = {"na...
Prologue: Git Hell "It's not that Git is hard. It's that we've made a mess, and now Git is telling us exactly how big a mess we've made." — Sarah, Lead Developer Three monitors, seven terminal windows, and a whiteboard full of hastily scrawled bran...
or: The Emperor Has No Clothes
Note: This blog post was inspired from Johannes Shickling’s tweet Your phone buzzes at 2:17 AM. The production monitoring system has detected a critical failure in your payment processing service. Heart pounding, you open your laptop to investigate....
