An applied AI lab
We make firms
AI native.
Granular builds AI systems in some of the most consequential environments on earth. We work at the level of the firm's infrastructure, records, and operations, turning frontier models into systems organizations can actually run on.
The systems run on Engram, a memory layer we build from everything the firm knows.
Deployments
Selected engagements.
National election campaign
The first political campaign in EMEA to deploy AI agents: 27 million voters, each targeted individually. The parties served were elected to parliament, one winning seats for the first time in its history.
$5 billion crypto fund
Research and diligence systems for one of the earliest and largest firms in the asset class, over its deal flow, communications, and market intelligence.
Global private equity firm
An always-on analyst embedded in the firm's daily operation. It reads the pipeline as it moves, prepares the partners for every meeting from the firm's own history with each person in the room, and has diligence underway before anyone asks for it.
Venture capital firm
The firm's entire written history, every email, file, and workspace page, compiled into one queryable memory. Research agents answer from it in seconds, with a citation for every claim.
Engram
The memory layer.
A firm's knowledge sits scattered across email, files, and meeting transcripts. Engram gathers it into one secure memory, and the systems we deploy draw on it to work the way an experienced member of the firm would. We believe it is the most capable memory layer in production anywhere.
What it absorbs
Email, files, meeting transcripts, and internal tools, through read-only connections. Every version of every document is kept.
What it builds
A knowledge graph of the firm's people, companies, deals, and decisions. Every fact is dated, traceable to its source, and replaced when a newer document supersedes it.
What the systems receive
Cited evidence, the current state of each claim, and any conflicts between sources, filtered by what the person asking is permitted to see.
How we work
The work is in the deployment.
Nearly every company now has access to the same few frontier models. The scarce ingredient is the execution work between a promising demonstration and a system a company can depend on: connecting it to real data, testing it against real cases, and making it behave acceptably on the bad days as well as the good ones. That is the work we do.
We work inside the client's firm
We send senior engineers and researchers to work with the client's data, on the client's infrastructure, alongside the people who will own the system afterward. We have tried other arrangements, and the systems they produced rarely survived contact with production, so we stopped offering them.
Every system is tested before it is trusted
Before anything goes into production we write an evaluation for it: a set of tests drawn from the client's actual work, which the system must pass and keep passing as it changes. We report the results as they come, including the ones that disappoint us.
The client ends up in control
The client's own staff finish the engagement able to run the system, modify it, and judge for themselves whether it is still working. A consultant who must be retained forever has, in our view, built the wrong thing.
Terms of engagement
Your data belongs to you.
Engram runs against the client's systems through read-only connections. Each firm's memory is encrypted, isolated, and remains its property.
The team that scopes it builds it.
You work with the same people from the first design conversation through production. We do not hand the work down or swap the team out.
We operate what we build.
A system has to hold up under real volume, real users, and a real quarter-end. We judge the work by how it behaves once people depend on it.
Inquiries
Write to us with the problem, the data it lives in, and who owns it.
inquiries@granular.workWritten inquiries are answered in the order received.