-
Chat Modes: Configuring the LLM for Each Turn
A real job for Jev: routing every chat turn to a mode with its own instructions, tools, model, and thinking.
-
Sequence Is Not Structure: Getting Lost in Long LLM Conversations
Long LLM conversations preserve sequence, not structure. I want a way to see where ideas branch, drift, and remain unfinished.
-
Moving Active Storage Between Buckets Without 404s
A small fallback service and rclone made it possible to move Active Storage files to a new S3 bucket without downtime.
-
Generative UI in Rails with RubyLLM
Let the model compose responses from a catalog of UI components. Rails validates and renders them.
-
Share Extension Auth in iOS 18: Four Approaches Compared
Share Extensions can no longer open your app directly. A comparison of four approaches to handle authentication in iOS 18.
-
Dictionary-Quality Word Pronunciation Without Dictionary APIs
Using LLM-based TTS for a language-learning app, with IPA to guide the pronunciation of ambiguous words like ‘read’.
-
Diving into Fizzy's Routes: Rails' resolve and direct
How Fizzy uses direct and resolve for public board links and models without their own pages.
-
GitHub Activity by Country
Commits per user by country, from the GHTorrent dataset in BigQuery and a small Ruby script.
-
Subdomain Routes in Rails 4.2+
Upgrading a multi-subdomain site from Rails 3.2 to 4.2: the old url_for hack goes away, and a small patch keeps the region: option.
-
Where Does Rails Keep CSRF Tokens?
CSRF tokens live in the session; since Rails 4.2 the token on the page changes on every load, but it is only the session token masked with XOR.