There is a particular failure mode we have both watched happen, from opposite sides.
On the research side: a result gets a number, the number goes in a paper, the paper gets cited, and the thing itself never runs anywhere that matters. The code is a notebook with a hardcoded path to someone's home directory. It works on the machine it was written on and nowhere else.
On the engineering side: a team builds something competent on top of an idea they did not develop and do not fully understand. When it breaks in a way the tutorial did not cover, there is nobody who can reason from first principles about why.
Both failures have the same shape. They are handoff failures.
The gap is not a communication problem
The usual prescription is better communication. Have the researchers write clearer documentation. Have the engineers attend the reading group.
We do not think that works, because the gap is not informational. It is a difference in what each side considers done.
A research result is done when it is defensible. An engineering artefact is done when it is operable. Those are genuinely different finish lines, and a handoff between them is a translation between two definitions of success. Translation loses things.
What we do instead
Do not hand off.
The person who trains the model is the person who deploys it. Not because that person is better at both — they are usually worse at one — but because there is no interface to lose information across. The trade-off is real: you get less specialisation and a lower ceiling on any single discipline. You get, in exchange, a system where nobody can say "that is the other team's problem", because there is no other team.
This only works at small scale. Three people. Maybe six. Past that you need structure, and structure means handoffs, and we are back where we started.
So we are staying small on purpose. That is the whole strategy, and it fits in one letter.
The letter
res + n + devs.
The n reads as and: Research and Development, plainly, both given equal weight.
It also reads as nexus — the hub where data, architecture, and logic converge. In computer science that word is precise, not decorative. It names the point where things meet and become a system.
That is the company. It is a conjunction, not a noun.