Patterns that hold

We build the software we advise on.

When a delivery pattern holds, we run the resulting software ourselves and install it with the operators who own the work — on their stack. Knowledge graphs and document intelligence are two of those systems.

Productised IP

Document Intelligence.

Every enterprise document becomes intelligence — captured, OCR'd and linked into a private knowledge graph, RAG-ready, with the source attached. Arabic and English, in-region by default.

PIPELINE · doc.ingest · type=invoice
OK

Invoice in. Payment status out — with the source page attached.

ASK

Is invoice #4471 overdue — and which contract does it bill?

ANSWER

Yes. Invoice #4471 is 47 days overdue (AED 186,400). It bills Contract #204 for Lex Group.

Source · inv-4471.pdf p.1 · linked CTR-204

Fields become nodes. The overdue flag is not a guess — it is a path through the invoice, the contract, and the customer.

At scale → a million pages ingested once — then every question searches all of them, in seconds.

Ontology & knowledge graphs

Your business, as a map.

Databases store rows. A knowledge graph stores meaning — entities, relationships, and the ontology that keeps them consistent so AI can reason with a source path attached.

QUERY · which renewals are at risk?
1 CRIT · 2 WARN
OWNSSIGNEDSIGNEDRENEWEDSIGNEDOVERDUEEXPIRESFAILEDBREACHEDESCALATEDAccount managerPerson · AM-0841CustomerLex GroupContract #310HealthyContract #204At riskRenewal date15 Mar 2027Invoice #4471Overdue +47dExpiry30 Sep 2026Contract #091CriticalPaymentFailedService promiseBreachedLegal deskEscalated

Ask in plain language; the answer is a visible path through your business — not a model guess. Gold traces a renewal warning; red is a payment failure that breached the service promise and reached legal.

Why it matters → LLMs guess from text. Graphs know from structure — grounded, multi-hop answers auditors can follow.

Start here

Book the AI Readiness Assessment

Fixed-fee. It pays for itself by killing the projects that wouldn't have worked — and scoring the ones that will.