Insights & Blog
Field notes from live deployments.
Written by the engineers on the SOW — not marketing. Short, opinionated, Kingdom-first.
AGENTIC AIJul 2026
Why 'agentic' broke under audit — and the middleware that fixed it
Role-scoped memory, tool allow-lists, and structured audit trails: what actually shipped after two banking pilots.
Read the note
ARABIC NLPJun 2026
Dialect coverage matters more than model size
Gulf, Levantine, and Egyptian dialect coverage moved the intent-classification F1 more than swapping model families.
Read the note
SPATIALJun 2026
The paper site map is a $6M SaaS bill in disguise
How three real-estate operators retired hosted-map dependencies with edge-rendered vector tiles.
Read the note
VISION 2030May 2026
PDPL is the easy part. NDMO is where projects die.
A pragmatic mapping of NDMO controls to real data-lake operations across the Kingdom.
Read the note
VOICE AIApr 2026
Bilingual voice agents: latency, dialect, and handoff
The three variables that decide whether a voice-AI worker actually replaces the qualification desk.
Read the note
MLOPSMar 2026
Evaluation harness > model choice
A minimal evaluation stack that survived four production LLM swaps without regression drama.
Read the note
Want these in your inbox?
One field note per fortnight. No product marketing.