Luke Chadwick / Berlin / Est. 2019
Twenty-five years shipping emerging technology before it was obvious. VR in 2014. AR in 2016. Production LLM systems since 2022. I work with engineering teams that need someone who has already built the thing they are about to build.
Shipped for
Scout24 · AutoScout24 · REA Group · LuxuryEscapes · Partmax · Zolidar
01 / Consulting
Fractional CTO, architecture audits, rescue work, and team enablement. Small, senior, hands-on engagements — no delivery pyramid underneath me.
See the engagements02 / Product studio
An AI-native property portal and multi-agent development tooling. Built in the open, funded by the consulting. This is where the methods come from.
See the productsAvailability
November 2026
One day per week — open
January 2027
Two days per week — open
Now
Mid-contract with two clients
Engagements
All of them are me, working directly with your engineers. I keep the footprint small on purpose — one or two days a week buys you the judgement, not the headcount. Scope and price agreed before anything starts.
01
Two weeks · fixed scope · fixed price
I read your codebase, interview your engineers, and give you a written assessment of what your AI systems will do under real load — plus a sequenced roadmap of what to fix first. The cheapest way to find out whether we should work together at all.
For teams who have shipped an LLM feature and are not sure what happens next.
02
One to two days per week · rolling
Architecture decisions, hiring, shipping cadence, and the technical calls a founder should not be making alone. I stay hands-on in the codebase — the leadership is more useful when it is grounded in the actual system.
For seed to Series B teams without a senior technical voice in the room.
03
Scoped per situation · starts within a week
The AI project that stalled, the vendor that walked, the prototype that will not survive production. I have done this before: in 2019 I rebuilt an automotive AR capture pipeline from scratch in four months after the vendor failed mid-contract.
For teams with a deadline and a system that will not make it.
04
Two to five days · on site or remote
Workshops that leave your engineers materially faster with AI coding agents — not a talk, a working session on your repo. Multi-agent workflows, review discipline, and where the failure modes actually are.
For engineering orgs where AI adoption is uneven and nobody owns it.
05
A few hours a month · monthly
A standing call and an open channel. Architecture reviews before you commit, a second opinion on the build-or-buy, and someone to escalate to when something breaks at 4pm on a Friday.
For teams who have the engineers and need the judgement.
How I work
Portrait of Luke — working, not posed.
A stranger needs a face before a call.
Twenty-five years in and I have never stopped shipping. Recommendations I cannot implement myself are not recommendations, they are opinions.
The measure of the work is whether your team can run the system without me. I am not building a dependency, and there is no account manager waiting to upsell you.
One or two days a week, one or two clients at a time. You get the person you met on the call, for the whole engagement.
Based in Berlin, delivered from 58 countries so far. On site when it earns its travel, remote when it does not.
Selected work
The pattern is the same every time: find the capability before it is obvious, build something real with it, hand it over. Here is what that looked like.
2022 — 2025
Production LLM systems
Problem. Strong engineering teams wanted to ship LLM features and had no one who had taken one to production before.
Approach. Embedded with the teams to build RAG pipelines, LLM integrations and agent workflows — then handed them over and left.
Outcome
Production systems running across EU, APAC and US — owned and extended by the client teams.
2014 — 2017
Co-founded the innovation lab. Built the Matterport partnership and a camera lending programme, then measured everything, in an industry that did not believe 3D tours changed buyer behaviour.
+95%
buyer enquiries
+140%
phone reveals
2019
The vendor failed mid-contract with millions committed. Rebuilt the Unity-based Android and iOS capture pipeline from scratch, under the original deadline.
4 mo
to shipped
1
contract rescued
The product studio
Two products that could not have existed two years ago. They are why the consulting advice is current: I am hitting the same problems your team is, on my own time and my own money.
In development
Not AI features bolted onto a legacy platform. Conversational search that understands tradeoffs, matching on what actually matters to a buyer, and agents handling the work that currently burns hours a day.
Real estate is the testbed. The architecture deploys anywhere.
In development
Several AI coding agents shipping to one repo without breaking each other. LLM-based priority arbitration, an automated merge queue, and human review pitched at the level where it still adds something.
The hard part was never one agent writing code. It is several of them shipping together.
The lab
Experiments I run to find out how something actually behaves before I recommend it to anyone. Some are infrastructure. Some are two novels turned into musicals. All of them are public, and most of them failed in an interesting way first.
Running · Infrastructure
Self-healing CI/CD. When the pipeline breaks, an agent diagnoses the failure, writes the fix and opens a pull request that merges itself if the checks pass.
Open experimentResearch hub · 200+ notes
A working encyclopedia on AI replicas, digital resurrection and what identity means when it can be copied. Two years of notes, cross-linked, still growing.
Open experimentPlayable
A ten-scene game that teaches prompt engineering by making you write the prompts. Graded live by a model, with a hall of fame.
Open experimentGenerative · Suno
Gibson's Pattern Recognition and Stephenson's The Diamond Age, adapted into full song cycles — an honest test of generative audio.
Open experimentBuild log
Ingestion plumbing for enterprise AI — Twirrl, tRPC and the unglamorous middle of every RAG system that actually works.
Open experimentAlso in here
DTOUR → Twirrl lineage · realestateVR field research · the Vibe Engineering quiz · Lab Notes, written from wherever I am
Thirty minutes, no deck, no discovery process. Describe the system and the deadline and I will tell you straight whether I am the right person — and if I am not, usually who is.
Response
Within one business day
Based
Berlin, EU hours
Next opening
November 2026