OIDC Won’t Stop the Next LiteLLM-Type Attack, but It Will Limit the Blast Radius
The LiteLLM compromise shows why CI/CD secrets are dangerous. OIDC cannot stop malicious code, but it can dramatically reduce what attackers keep.
Featured writing
Featured article
Resume
Senior full-stack engineer with 11+ years of experience in .NET, JVM, microservices, production AI systems, AI-assisted delivery, and developer tooling.
Recent writing
Notes from the workbench
Secure C# Starts with System Design
Secure C# begins with system boundaries and disciplined use of the language: narrow APIs, immutable state, short-lived secrets and deliberate dependencies.
Don’t Just Plug Holes in Mutation Tests
A surviving mutant is not automatically a request for another test. It may be exposing unnecessary code, a missing specification, or a design responsibility in the wrong place.
Make VS Code open Markdown in preview by default
A tiny VS Code workspace setting that makes Markdown files open in preview mode instead of the editor.
Boring as a Feature
Boring code is not a failure of imagination. It is often the most maintainable choice, with cleverness reserved for experiments that earn their cost.
Building a Cheap Static Website with Astro and GitHub Pages
A practical beginner-friendly guide to building a polished personal website cheaply with Astro, GitHub Pages, and a custom domain.
Selected public projects
A short portfolio of public projects covering AI tooling, semantic search, C# engineering, frontend work, VR, Rust, and local-LLM experiments.
Tome: semantic search for books, docs, and source code
Tome is a Python MCP server for local semantic search over books, long-form documents, and C# source files using Ollama embeddings and LanceDB.
Building a custom job-search ranking workflow
How I used an AI-assisted browser workflow to turn noisy job listings into a ranked, reviewable shortlist.
AI needs guardrails, not rubber stamps
Approving AI-written code is too easy to rubber-stamp. The quality comes back when you write the scaffolding, tests, tooling, and constraints first.
My resume is now a static site feature
I turned my resume into a versioned, printable, mobile-friendly page that is easier to share and keep current.
Going meta: making AI think like me
How I use Claude Code to keep myself on track — PRDs for scope, pomodoros for pace, rules files so I don't have to hold everything in my head.
I built a /blogpost command that writes, illustrates, and publishes for me
One /blogpost command: Claude drafts the article, generates cartoon illustrations with Cloudflare's free AI tier, previews in the browser, publishes to GitHub Pages, and fires off a LinkedIn post — all without leaving the terminal.
Claude Code is wasting ~9,400 tokens every session — here's the fix
Claude Code silently loads MCP server definitions for integrations you've never used. Here's how to find out how bad it is and fix it in 30 seconds.
Running a 1-Bit LLM on Windows — Bonsai 8B Setup Guide
An 8B LLM running at 180+ tokens/sec on a laptop GPU using just 1.1 GB of VRAM — complete walkthrough for Bonsai 8B on Windows with CUDA.
Avoid death by accountant - Get bankers rounding right!
Why bankers rounding is the right default for financial code in C#, how it works, and how to avoid an angry accountant when ledgers don't balance.
Topics