Blog
Build Log · how we actually run D3X
Replacing Alexa with a Home Assistant Local Voice Assistant
Our first-hand build of a bilingual Home Assistant local voice assistant: what runs on the box, what we sent to the cloud, what broke, what it costs, and why microphones are the real obstacle.
Zigbee vs Thread vs Wi-Fi: Picking a Reliable Smart-Home Network
Lessons from a one-week Home Assistant build: why we put 22 of 24 lights on Zigbee, kept Wi-Fi for media and metering, and deliberately parked Thread until the host could support it properly.
We Stopped Renting Our Own Photos: 62,157 Memories Moved from Google Photos to Immich
Google Photos is free until your memories pass 15 GB. Ours were 565 GB. Here is how we moved 58,039 photos and 4,118 videos to Immich on our own NAS, what the Takeout tried to do to us, and the one thing Google really was doing for the money.
Latest articles
Self-Hosted LLM on Your Own GPU Box: What It Really Takes
Running a self-hosted LLM takes a correctly sized GPU, an efficient inference server and an operational layer for auth, monitoring, updates and fallback, and that last part decides whether it survives.
Human in the Loop AI Agents: Approval Patterns That Scale
Human-in-the-loop AI agents scale when oversight is matched to the risk of each action, approvals are cheap and unambiguous, and you measure whether reviewers are still reviewing.
AI Proof of Concept to Production: Why Most Pilots Stall
Most AI pilots stall because the proof of concept tested whether a model could do the task, not whether the business could run it every day, so here is how to design a pilot that is built to reach production.
Dataflow Security in AI Pipelines: Map, Restrict, Audit
Most AI data leaks happen in the plumbing around the model, so dataflow security means mapping every hop, restricting each one and auditing until the map stays true.
Self-Hosted LLM vs API: Cost, Latency and GDPR Trade-offs
For most businesses an LLM API wins on cost and speed to ship, and self-hosting an open-weight model only pays off with high steady volume, strict data control or tight latency needs.
RAG vs Fine-Tuning: Which Does Your Use Case Actually Need?
For most businesses the answer is retrieval first: RAG fixes what a model doesn’t know, fine-tuning changes how it behaves, and plenty of use cases need neither.
AI Invoice Processing Automation: A Worked Example
A step-by-step worked example of AI invoice processing automation, from the AP inbox to a draft bill, showing where the LLM fits, where plain code must take over and where a human still decides.
Should I Automate My Bottleneck First? Not Always
Automating the bottleneck first is usually right, but only once it is confirmed as the real constraint, its cheap fixes are exhausted, and the work there is rules rather than judgement.
LangGraph vs CrewAI vs Custom: Choosing an Agent Framework
Choose LangGraph for explicit state and control, CrewAI for fast role-based prototypes, and custom code when your agent is one model calling a few tools, but draw the workflow before you pick any of them.
We Stopped Renting Our Own Photos: 62,157 Memories Moved from Google Photos to Immich
Google Photos is free until your memories pass 15 GB. Ours were 565 GB. Here is how we moved 58,039 photos and 4,118 videos to Immich on our own NAS, what the Takeout tried to do to us, and the one thing Google really was doing for the money.
My Wife’s Claude Code Talks to My Agents: Two Claude Accounts, One Home Lab, Six Locks in Between
My wife has her own Claude subscription. Now she has her own Claude Code session too, on her account, opened from her phone. It can message me, run everyday things in the house and ask my agents to find things. It can’t touch the lab. Here is how it’s wired, and the two bugs that hid my replies.
An Unattended Windows 10 USB Stick, Built on a Mac: Wipe, Install, Drivers and Apps in One Boot
We rebuilt a family member’s ThinkPad T14s from a USB stick made on a Mac: plug in, press F12, walk away. Polish Windows 10 Pro, Lenovo’s drivers, no bloat, five apps. Here is every script, and every way it broke first.
Home Lab Documentation That Keeps Itself Current: Gitea, GitHub and a Nightly Snapshot
Every repo in our lab lives on a self-hosted Gitea first and is mirrored to private GitHub as a backup. A wiki rebuilds itself minutes after each change, and every night the live configs of the lab are copied into git with the secrets taken out and the inventory pages are rebuilt from the live systems.
Fractional CTO for AI Projects: When It Makes Sense
A fractional CTO for AI earns its fee when your business faces AI decisions that are expensive to reverse but lacks the continuous senior technical work to justify a full-time hire.
Enterprise Ireland AI Grant: What Funding Can You Actually Get?
There is no dedicated Enterprise Ireland AI grant, but several digitalisation and innovation supports can part-fund AI work if you pick the one that matches your company and your project.
What We Do
Product
Consumer and enterprise software, built with taste and shipped fast.
Infrastructure
AI agents, automation, and orchestration - production systems, not demos.
Advisory
Hands-on engineering and strategy for teams shipping AI into production.