~/mk / uses

/uses

The exact hardware, software, and data stack I use day-to-day. Transparent, no affiliate links, no "sponsored by" rot. Inspired by usesthis.com and every developer who's written one before me.

last updated april 2026
// 01

hardware

Keeping it modest until my first full-time contract. I edit, compute, and render on the same machine.

  • MacBook Air M2 13", 16GB / 512GB. Silent, fanless, runs models locally until they don't fit. daily driver
  • Dell U2723QE 27" 4K IPS. Single-cable hub into the laptop. Enough real estate for SQL on the left, Power BI on the right. display
  • Keychron K3 Pro Low-profile, brown switches. Quiet enough for co-working, tactile enough to stay honest. keyboard
  • Logitech MX Master 3S Horizontal scroll earns its keep in any spreadsheet wider than the monitor. mouse
  • Sony WH-1000XM5 Library hours, Brussels trams, and pair-debugging calls. headphones
  • Moleskine Cahier + Muji 0.38 Every project starts on paper. Boxes, arrows, questions. Then code. analog
// 02

editor & dev

I try to keep the toolchain boring so the problem can be interesting.

  • VS Code Night Owl theme, JetBrains Mono at 13.5px, 1.6 line-height. Vim keybindings extension for light-touch motion. editor
  • Claude Code Long-context pair for refactors, notebook cleanup, and rubber-ducking analysis. copilot
  • iTerm2 + zsh + starship Minimal prompt, branch + env indicators only. No emoji. terminal
  • uv Replaced pip + venv + pip-tools in a single binary. Project installs in seconds. python pkg
  • ruff + black Ruff for lint, black for format, pre-commit to enforce. No opinions beyond the defaults. lint/format
  • GitHub + gh CLI Issues first, branches per problem, squash-merge. gh pr create from the terminal. vcs
// 03

data stack

The actual tools that touch client data. Picked for reproducibility over novelty.

  • Power BI Desktop Primary BI canvas. Semantic models, DAX, Tabular Editor 3 for anything serious. bi
  • DAX Studio Non-negotiable for query profiling and VertiPaq analysis. Catches bad measures before production does. bi
  • DuckDB My go-to local query engine. CSV, Parquet, even Excel — SQL over everything in-process. query
  • Polars Moved most notebook wrangling off pandas. Predictable memory, predictable speed. dataframe
  • scikit-learn + XGBoost 90% of models I ship. Boring, interpretable, fast to re-train. ml
  • SHAP If a model has business impact, it needs explanations stakeholders can read. ml
  • Streamlit Zero-to-prototype for internal tools. Upgrades to a real frontend only when it's earned one. app
  • dbt-core For any project with more than one join. Models, tests, docs — in that order. transform
// 04

ai & agents

The stack behind MatterMind and most of my side experiments.

  • Claude (Sonnet & Haiku) Long-context reasoning and tool use. Haiku for cheap, fast ranking passes. llm
  • LangChain + LangGraph Agent orchestration when a single prompt isn't enough. Used sparingly — most problems want a function, not a graph. framework
  • FAISS Embeddings index for research-paper retrieval. Local, free, fast enough. vector
  • Model Context Protocol Writing MCP servers to give agents first-class access to my own data tools. protocol
// 05

cloud & infra

Small surface area on purpose — I don't want to pay for something I'm not learning from.

  • GCP BigQuery for scratch warehouses, Cloud Run for Streamlit deploys, GCS for everything in between. cloud
  • Vercel This site, plus any landing pages or docs I ship. Zero-config wins. hosting
  • Docker Slim Python images, multi-stage builds. Reproducibility first, image size second. container
  • 1Password Secrets in the CLI via op run. No more .env files in the home directory. secrets
// 06

writing & thinking

The notebook layer between raw notes and anything a stakeholder sees.

  • Obsidian Vault synced to a private git repo. Daily notes, project logs, paper highlights. notes
  • Excalidraw System diagrams and data flows. Ugly-on-purpose visuals beat pretty-on-purpose ones. diagrams
  • Figma Dashboard wireframes before I touch Power BI. Saves more time than it costs. design
  • Things 3 GTD with a light hand. Inbox, today, projects — nothing more elaborate. tasks
// 07

desk & audio

What the chair looks like when deep work is actually happening.

  • Flexispot E7 Standing desk, used ~30% standing. The budget frame with a plain oak top. desk
  • IKEA Markus Honest chair. Not fancy, not broken, for the full price of a month of transit. chair
  • Endel · brain.fm · Nils Frahm Deep-focus rotation. Lyrics during admin, no lyrics during code. focus
  • Bialetti 3-cup Moka pot on a small induction ring. Two per day, third on deadline nights. fuel

Nothing on this page is sponsored. If something works for me, it's here. If it stops working, it leaves. Swap suggestions are always welcome — send them to [email protected].

← back to home