Digital product · AI · Design & build
Sorting Hat
An intelligence layer over a LinkedIn network of 10,000 contacts. Every contact sorted into groups, the few worth a conversation surfaced first, and what is new about them delivered with sources. Shown here on invented sample data.
The work
Every claim arrives
with its source.
Sorting Hat reads a LinkedIn connections export and sorts every contact into groups that mean something: funds, founders, operators, press. A scoring pass promotes the few people worth attention first, and a research pipeline watches them, surfacing what has changed: a fund closed, a role moved, a company sold.
The rule that shapes the whole build: an AI claim without a source does not exist. Every finding links to where it came from, and anything the pipeline cannot source is refused rather than guessed. The rules are enforced in the database itself, and the logic is tested against fixtures, never against a live model.
There is no public demo, and that is deliberate. The product holds a real professional network, so it lives behind a login, reads LinkedIn without the ability to write back, and shows its data to exactly one person. The screenshot above is the product running on invented sample contacts. This is how we build for businesses whose data cannot be public.
Classification
10,000 contacts sorted into groups that mean something, with a review queue for the uncertain ones.
Prioritisation
Scoring that promotes the few people worth a conversation, so attention goes where it pays.
Sourced research
Live findings on the people who matter, every one linked to where it came from.
Private by design
Login-only, read-only towards LinkedIn, one reader. No public demo.
Next project
Biboja
Visual identity · Website