Research
Some questions are not build questions yet. We take on research collaborations: the literature review, the feasibility study, and the unglamorous work of finding out whether the thing you want has already been tried and why it failed.
What it is
We join a research endeavour as the AI half of the team. A university group, a product team or a founder brings the domain and the question; we bring the models, the evaluation, and the discipline of writing down in advance what would count as an answer. You get a result you can cite, not a deck that agrees with you.
The research desk is where the credibility for that comes from. It is a public, dated log of the papers and benchmarks behind how we build, each credited to its real authors and annotated with what it changes in production. It is the reference layer we argue from, and it is public so our answers can be checked.
How we build it
Agree what would count as an answer
Before any reading, we write the question down in a form that could come back false. Most research engagements go wrong here: the question is really a request for confirmation, and no amount of literature fixes that.
Read the field, including the failures
We go to the primary sources and look specifically for what has been tried and abandoned. A result nobody reproduced is different evidence from one three groups reproduced, and the negative results are usually the ones that save a budget.
Deliver something citable
You get the sources, the method, what we found and what we could not establish. When the honest answer is that the evidence does not support the plan, that is the deliverable, and it is the most valuable one we produce.
Who it is for
A good fit if
- Teams with a research question and no machine learning capacity in house.
- Founders and product teams who need to know whether an idea is feasible before they fund it.
- Academic and institutional groups who want an AI collaborator on a defined piece of work.
- Anyone evaluating a model or a vendor on claims they cannot verify.
Not a fit if
- Work that needs a literature review to reach a conclusion already agreed.
- Requests for a citation that supports a decision you have already made.
- Open-ended exploration with no question attached. That is a retainer, not research.
See it
A reference build
What a collaboration looks like
A defined question, a scoped read of the field, and a written answer with its sources attached. We do not run these as open-ended exploration. The question is agreed first, in a form that could come back false, and the deliverable states plainly what the evidence supports and what it does not.
- A question and a method agreed in writing before any reading starts
- Primary sources with the negative results included, not only the wins
- A deliverable that says what could not be established, and why
This is a system we run, not a client case study. We do not publish client numbers we cannot show you the workings for.
Talk to someone who builds this
No qualification call, no discovery deck. Describe the work and we will tell you whether this is the right line for it.