Articles
Write-ups from the work. MCP servers, AI workflows, content engineering, search, and the occasional tangent. Most of these come from problems I hit on real engagements, written up so the next person does not have to figure it out from scratch.
-
· local-llm · vllmThe VRAM traps: why a 16GB model wouldn't load on a 48GB card
A 16GB model would not load on my 48GB card, and the reason was a shortfall of one kilobyte in a memory no spec sheet mentions. These are the fit traps a VRAM figure will never warn you about.
Read article -
The Dual-4090 96GB vLLM Benchmark & Runbook
Can a £6,200 modified Ada rig match enterprise MoE throughput? The living measurement record for a dual RTX 4090 48GB vLLM rig - every number measured here.
Read article -
· local-llm · vllmThe broken rules of local LLM inference
I used to lock the clocks on my mining GPUs. The same instinct just helped kill five rules of local LLM inference on a £6,200, 96GB rig.
Read article -
· Claude · mcpHow to Stop MCP Servers Eating Your PC: One Docker Gateway for Every Claude Client
My MCP list grew until orphaned node.exe were quietly eating a 128GB workstation by mid-afternoon. Here's how I put every server behind one Docker gateway - node on bare metal gone, secrets in one gitignored file, and a single URL every Claude client points at. One evening's work you'll feel every day after.
Read article -
· Claude · mcpHow to Use the Gemini API (and Why I Run It Next to Claude)
Get a Gemini API key, make your first call in curl and Python, dodge the thinking-token trap that returns an empty answer, and see why running Gemini next to Claude is the real unlock. Written from production - and the bills.
Read article -
Claude Desktop Makes a Brilliant Coding Assistant - Here's How to Set It Up
Everyone says if you want AI coding help, you use Cursor. Or Cline. Or GitHub Copilot. Basically anything that hooks into your IDE with inline autocomplete and tab-completion magic. Claude Desktop? That's for chat, not code. I respectfully…
Read article -
· Claude · geminiWhich AI is right for your job? ChatGPT, Claude, Gemini and Copilot, task by task
Everyone's talking about AI assistants - ChatGPT, Claude, Gemini, Copilot - and if you've a real job to get through, you mostly want to know one thing: which of them will build Monday's deck, sort the badly-exported spreadsheet, write up the meeting, and clear the inbox? Here's the answer, task by task.
Read article -
· gpu · hardwareMuse Glimmer 30B vs qwen: my day-one local benchmark on a dual RTX 4090 rig
Meta dropped Muse Glimmer 30B and I spent the day trying to unseat my qwen daily driver on the dual-4090 rig. Four apparent hangs, one day-one bug, and a same-harness bench later, I had my answer - and it wasn't the coronation I'd half expected. So, is it better than my daily driver?
Read article -
· claude-code · AI codingHow to Plan and Begin Your First AI-Assisted Coding Session
You don't need to know how to code to build something with AI - but the calm ten minutes you spend planning before you start is what keeps your first session from spiralling. Here's the whole thing, gently: what the tools are in 2026, how to plan, and exactly what your first session looks like.
Read article -
· claude-code · PRDHow to Write a PRD an AI Can Build From (with a template)
A PRD is the difference between an AI coding tool that guesses and one that builds the thing you meant. Here's what a PRD is, a copyable seven-part template, a worked example, and the two lines that do most of the work - written for the era where the thing reading your spec is an agent, not just your engineering team.
Read article -
· claude-code · mcpHow I work with Claude Code: PRD to deploy
I write the PRD, we agree an execution plan, pick the environment, get the keys out of the way, prototype, then test. Six steps, same order every time. Here's the real workflow with the files from two things I've shipped, and which model I hand each job to.
Read article -
· claude-code · local-llmThe best local coding setup isn't one model: how I route across Claude, Kimi and my own rig
The question I get asked is which local model is best for coding. Wrong question. The setup that works routes three tiers - Claude reasons, Kimi builds, and a Qwen coder on my own rig does the volume for nothing. Here's the whole thing, wired up.
Read article -
· local-llm · vllmHow to set up vLLM in Docker: serve an open-weight model on your own GPU
vLLM in Docker, from empty machine to an OpenAI-compatible endpoint you can curl: the compose file I run, the flags that survived benchmarking, and the deadlock that hides behind a healthy /health check.
Read article -
· claude-code · mcpMoving houtini-lm to vLLM: What I learned
I decommissioned Hopper (my local LLM bootstrapped server) and moved my local models to a two-card 4090 rig with vLLM on Docker. It's so much faster - but houtini-lm spat its dummy. Two bugs, one hiding behind the other, and how v3.2.1 fixes it.
Read article -
· Claude · claude-codeFrom Chat to Deploy: How to Fully Understand Claude, AI Assistants and Claude Code
I started in the chat box like everyone else. Three years later the boring parts of my work run themselves, from PRD to deploy. This is the ladder, stage by stage, with a starter task for each rung.
Read article -
· mcp · claude-desktopHow to Do a Technical SEO Audit with Claude
A free, step-by-step technical SEO audit with Claude: your Search Console history and a first-party crawl merged in one local database, ranked by recoverable clicks - the Screaming Frog alternative you run by conversation.
Read article -
· local-llm · gpuTuning vLLM on a Modded 48GB RTX 4090: From 18 to 133 Tokens per Second
Three days of benchmarking vLLM on a modded 48GB RTX 4090: the quantisation trap that costs Ada owners 40% of their speed, the free 1.9x from speculative decoding, the KV compression that never served once, and what cutting 120 watts costs you. Every number measured, every failure kept.
Read article -
· claude-code · mcpGiving Claude a Local Sidekick: a vLLM Journey Update
I want Claude to have a powerful local sidekick through houtini-lm. I'm not there yet, but this month I got noticeably closer - a vLLM rebuild, a benching spree on the 48GB 4090, and a face-off that deleted two models in an evening.
Read article -
· local-llm · gpuBest GPUs for Running Local LLMs (2026): VRAM, Bandwidth & Picks
What to buy in 2026 for local LLMs, benched first-hand: why memory bandwidth beats FLOPS, why VRAM decides everything, from the used 3090 to the modded 48GB 4090.
Read article -
· mcp · system-requirementsClaude Desktop System Requirements (2026): Minimum & Recommended Specs
The minimum and recommended specs to run Claude Desktop in 2026: what the official page claims, what real machines actually need (RAM, OS, chip), and where the install trips up.
Read article -
· claude-code · cliClaude Code System Requirements (2026): Specs, Setup, Gotchas
Claude Code on Mac, Windows and Linux (2026): the real-world spec, the install gotchas the top guides skip, and how to cut the token bill with a local LLM.
Read article -
· strategy · claude-codeAI for the CFO: variable cost, vendor sprawl, and the playbook that ends the surprise bill
Uber reportedly provisioned 5,000 engineers with Claude Code in December 2025. By April 2026 it had burned the entire annual AI budget in four months. Variable cost is the new shape of the AI line item, and the CFOs who are getting it under control are running the same four-lever playbook. Here is what is on each lever, which platforms are credible, and what to ship in the next ninety days.
Read article -
· Claude · strategyAI hallucination, and the boring discipline that stops it being a problem
You have read about lawyers citing made-up cases and chatbots inventing refund policies. The fix is not a new platform or a clever prompt. It is the most boring discipline in software: read what came out and check the bits that matter against a second source. Here is what hallucination is, how we run a Houtini-grade check on every claim, and what would change if your team did the same.
Read article -
· Claude · strategyWhat is RAG, and what could it do in your company?
Your AI confidently answers questions about your business with public-internet knowledge. RAG is the architecture that gets your actual contracts, customer list and operating playbook into the conversation, with citations. Here is what it is, where it sits next to long context and MCP, and what it changes if you sponsor the work this quarter.
Read article