May 1 · 8 min read · A practical guide to TypedDict, dataclass, and Pydantic for ML engineers Python is a duck-typed language. That means Python does not care what a value is. It only cares what a value does. If it walk
Join discussion
Apr 27 · 12 min read · In Part 1, we built the simplest possible thing — a Python script that fetches a failed GitLab job trace and asks Claude to classify it. It worked. But the moment it worked, we saw every crack clearly
Join discussionApr 17 · 11 min read · If your production codebase contains a function called try_parse_llm_response that runs re.search(r'\{.*?\}', response, re.DOTALL), you are not alone — and you are also in 2023. In 2026, parsing LLM output with regex is the equivalent of hand-rolling...
Join discussion
Apr 13 · 11 min read · Here's a class of bug I wrote in my first LLM app and then again, on a different team, six months later. I wrote a feature that asked the model to classify a user's message into one of three categories — billing, technical, other — and return it as J...
Join discussion
Apr 7 · 9 min read · What if an AI agent could recall information not just accurately, but with absolute structural integrity? Imagine an agent failing a critical task because a key piece of data was malformed, leading to catastrophic failure. This is precisely the probl...
Join discussionMar 27 · 7 min read · Product Finder is a retail data platform prototype built for SportsWear AB. The goal is simple: take inventory and sales events from operations, process them in a reliable pipeline, and turn them into
Join discussion
Mar 20 · 4 min read · import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem'; Simon Willison shared research on running Pydantic's Monty in WebAssembly. Monty is a minimal, secure Python interpreter written in Rust, designed specifically for safely executing...
Join discussion
Mar 10 · 9 min read · LangGraph gives you fine-grained control over your agent's execution graph — you define nodes, edges, and conditional routing. But one thing that's missing from most LangGraph tutorials is what happen
Join discussion
