Skip to content

The Lab

A lab, not an agency.

We build products. We take a few client engagements. Each one sharpens the other.

  • 2026 Founded
  • 6 Practice areas
  • 3 Engagements at a time

Embedded, not outsourced.

  • The products come first.

    We build and run our own systems across six practice areas. Every engagement is built by the same people, on the same stack, with what those products taught us.

  • The people you meet are the people who build it.

    Nothing is handed off. Whether it is a fixed-scope build, a review, or an engineer forward deployed inside your team, the people on the scoping call do the work, start to finish.

  • Three engagements at a time.

    A short list is what makes the first two possible. It keeps the product work moving and keeps every engagement senior.

Open questions

Not solved yet

These are the problems on our bench. If one of them is yours, we would rather compare notes than guess.

01

How do you evaluate an agent that is allowed to be creative?

Tests catch regressions in code. They say nothing about a run that took a different but valid path. We are scoring outcomes and constraints instead of transcripts.

Open question 01

How do you evaluate an agent that is allowed to be creative?

02

Where is the cost ceiling on verifiable game outcomes?

Proof generation is affordable for a final and far too expensive for a casual match. We are measuring where that line sits, per circuit and per chain.

Open question 02

Where is the cost ceiling on verifiable game outcomes?

03

Can retrieval stay correct as a corpus outgrows its authors?

Stale documents outrank current ones and nobody notices until an answer is confidently wrong. We are testing freshness policies that fail loudly instead.

Open question 03

Can retrieval stay correct as a corpus outgrows its authors?

Muhammad Abdullah, studio portrait in a charcoal polo shirt.

Muhammad Abdullah

Co-founder & Engineer

Four rules we do not bend.

We say no

If a problem is outside what we are good at, or the timeline is not real, you hear it on the scoping call. You lose thirty minutes, not three months.

If it teaches us nothing, we pass

Every engagement has to leave the lab knowing something it did not. Work that is only billable goes to someone else, however well it pays.

Three at a time, never a fourth

The cap is fixed. When the third seat is taken you wait, and so does everyone else. Nobody gets a diluted team because a bigger cheque arrived.

We publish what we have not solved

The open questions we publish are real and current. You can see what we are stuck on before you hire us, which is more than a case study will tell you.

If the problem has a hard core, we want to hear it.

Free consultation