Topic
#llm
3 items
Posts
- How magj.dev Grew Up
Three hundred commits is not a lot. It is also not a little. This is the story of how a personal Astro blog quietly became infrastructure — an SSR Lambda, a single-table DynamoDB Content API behind CloudFront, an admin surface with MFA, a streaming LLM proxy on a Function URL, and per-PR ephemeral staging — one decision at a time. Plus the things I shipped, the things I reverted, and what the architecture taught me along the way.
- One Gateway to Rule Them All
The sequel. Last month I shipped a Rust LLM proxy and called it a solid foundation. This month I found out what that actually means: migrating five projects through it, wiring AWS Bedrock, adding Gemini, and making a FinOps decision that cut baseline model costs by 97 percent.
- I Didn't Expect GPT-5.4 to Change My Coding Workflow. Then It Did.
I came in as a genuine Claude fan with a working Opus-plus-Sonnet workflow. GPT-5.4 with high reasoning still managed to impress me — not with raw code generation, but with better upfront planning, cleaner tradeoffs, and the ability to spot meaningful improvements over already-strong Opus output. Here is what changed and why.