- 01Applied ML4 minYour model’s 90% confidence interval is probably 85%Quantile regression produces intervals that are systematically too narrow. Here is how we measured the gap on a live valuation model, and the calibration layer that closed it.Read the post
- 02AI Engineering4 minOAuth 2.1 between a web app and an MCP serverConnecting a product to ChatGPT and Claude means running a real authorization server. What the flow looks like end to end, and why standards compliance beat per-vendor branches.Read the post
- 03Architecture4 minMulti-tenant isolation you can prove in a compliance reviewScoping data by tenant is easy to claim and hard to demonstrate. The two structural decisions that turn “we isolate tenants” into something you can answer with a query.Read the post
We publish when a project produces something worth writing down, not on a schedule. If a post raises a question about your own system, ask us directly.