Portfolio

/uses

What I use to ship.

Inspired by uses.tech. The exact hardware, editor setup, AI tools, and methodology I use day-to-day to build production AI. Updated as things change.

Hardware

  • MacBook Pro M3

    Daily driver. 16″, 36GB RAM.

  • External monitor — LG UltraFine 27″

    4K, vertical for code reviews.

  • Logitech MX Master 3S

    Worth every fil.

  • Keychron K2

    Brown switches.

Editor & coding

  • Cursor

    AI-first editor. Worth the subscription.

  • VS Code

    Backup. Same extensions where it matters.

  • GitHub Copilot

    On for boilerplate, off for architecture.

  • Theme: Tokyo Night Storm

    Easy on the eyes for long sessions.

  • Font: JetBrains Mono

    Ligatures on.

AI tools

  • Claude Sonnet 4 / Opus 4.5

    Primary AI pair-programmer.

  • ChatGPT (GPT-5)

    Second opinion + image generation.

  • Cursor agents

    Long-running refactors.

  • Perplexity

    Up-to-date research.

  • Groq Playground

    Latency-sensitive prototyping.

Terminal & shell

  • Warp

    AI-native terminal.

  • zsh + Oh My Zsh

    Powerlevel10k prompt.

  • lazygit

    Faster than the GUI.

  • ripgrep, fd, bat

    The unix toolkit, modernised.

Languages & frameworks

  • TypeScript

    Frontend + Node services.

  • C# / .NET 8

    15 years and still my favourite for backend.

  • Python

    ML / scripting / FastAPI.

  • Next.js 14

    App router. This site runs on it.

  • Tailwind CSS

    Productive when constrained.

AI / infra stack

  • Azure OpenAI

    Production GPT-4o for SCAD systems.

  • Groq

    Sub-second Llama 3.3 70B for low-latency UX.

  • Pinecone

    Serverless vector DB.

  • Cohere rerank

    Precision lift on retrieval.

  • LangChain / Semantic Kernel

    Orchestration when warranted.

Productivity

  • Notion

    Specs, notes, project tracking.

  • Obsidian

    Personal knowledge base.

  • Linear

    Issue tracker for solo & team projects.

  • Raycast

    Replaced Spotlight years ago.

  • Arc Browser

    Workspace per project.

Methodology

  • Evaluation harnesses, always

    Before models, before prompts.

  • Measure latency, accuracy, cost

    Three numbers, every release.

  • Two pizzas, one ticket

    If it can't fit, split it.

  • Boring tech wins

    Production AI on Azure, not bleeding-edge OSS.

Last updated · May 2026