Insights
Engineering perspectives on architecture, product development, AI, and the everyday decisions behind useful software.
From the engineering desk
Browse practical articles or follow the latest technology updates.
Subscribe via RSS →Curated links from external sources — not 360Softy original articles.
Your AI agent confidently tells a user that the company's parental leave policy is 12 weeks. It's been 16 for the past year. The old HR handbook, the updated one, and the Slack announcement that changed it are all sitting in the retrieval index. The a...
now supports .Vercel AI Gatewayrouting rules Routing rules are firewall-style rules that control which models your team can use, applied at the gateway level instead of in your application code. When a model goes down or gets retired, you usually have to ship a code change to move off it. With routing rules, you push one rule and every request reroutes instantly. There are two types: Rules apply to every request made with your team's AI Gateway credentials. You manage them with the .Vercel CLI C
Your origin can serve different responses for the same URL — different languages based on Accept-Language, or different formats based on Accept — by returning a Vary response header. Cloudflare's cache now honors that header directly in Cache Rules, so the same URL can hold multiple cached versions and each request is matched to the right one. Content that previously had to bypass cache to stay correct can now be cached, following standard HTTP caching behavior. What changed Your origin now deci
A subtle trap of taking a SaaS international: the system emails sent from your Stripe webhook. Purchase confirmation, renewal success, payment failure, plan change — four kinds of emails triggered by Stripe events. We recently discovered all four had been hardcoded to Japanese for months, sending Japanese receipts and failure notices to English-paying users overseas. The kind of bug that quietly persists forever unless you go looking. This post walks through the currency-based language inference
Comments
If you've ever had to merge a few PDFs, stamp a "CONFIDENTIAL" watermark, or turn a stack of images on the server, you know the pain: pdf-lib, pdftk, ghostscript, native Here's a lighter option: a small HTTP API that does the common PDF jobs for you. You POST a file, you get I'll show merge, watermark, and images-to-PDF with copy-paste examples in cURL and Node. The PDF Toolkit API exposes six routes: Endpoint Does POST /v1/pdf/merge Combine multiple PDFs into one POST /v1/pdf/split Spl
Mistral introduced a new version of its document-reading model as a hosted OCR service, and the open-source project MinerU has been climbing fast on GitHub doing a similar job you run yourself. Both aim to convert messy PDFs into clean, structured text that AI systems can actually use, and that progress matters because bad document reading silently caps the quality of everything built on top of it. What: Mistral released a new document-reading model the same week an open-source rival surged, bot
We spend a lot of time arguing about code quality. Is it clean? Readable? Tested? Too clever? AI-generated? All fair questions, but I think we're missing a more important one: Can the repository actually prove what this behavior is supposed to do? Not the developer. Not the product manager. Not the person who has been on the team for four years and somehow knows that this weird edge case is intentional. The repository itself. Because when a system gets old enough, "how it works" and "what the re
Let’s start with a conversation
An idea, a challenge, or a system that needs to work better. We’ll help you understand the next step.
Prefer email? [email protected]
