Audit AI · تدقيق الذكاء الاصطناعي — ICAIRE glossary and on-device AI compliance audits.

Audit AI
Arabic-first glossary & governance tools · 1,242 ICAIRE terms
⌘K

ICAIRE · AI governance

Audit AI

An Arabic-first glossary and compliance workspace.

Numbers at a glance

Curated from the shipped glossary, graph, rubric, and Hugging Face artifacts.

  • 1,242 ICAIRE glossary terms enriched
  • 12+ structured fields per term
  • 7 story tracks covering ~1,140 terms
  • ~1,180 auto-generated Mermaid UML diagrams
  • ~8,400 typed graph edges (prerequisites, unlocks, is-a, part-of, contrasts-with, used-with, alternative-to)
  • 519 governance controls extracted across 4 frameworks
  • 1024-dim BGE-M3 embeddings precomputed for both corpora
  • 6 export formats (master JSON, graph JSON, story map, GEXF, Cypher, TBX, Anki)
  • 3 Hugging Face datasets (rubric + embedded corpora); ICAIRE enriched glossary recommended
ICAIRE matrix — narrative

Who we are, what problem we solve, and what this workspace delivers—grounded in the Arabic-first glossary and governance rubrics.

Who we are

Purpose

We combine the ICAIRE Arabic-first glossary with tooling to review AI-related documents against recognized governance frameworks—so teams can align language and evidence in one place.

The challenge

What we are dealing with

Policies and project documents are fragmented across languages; compliance expectations differ by OECD, UNESCO, and sector-specific controls. Teams need searchable terminology, traceable evidence, and repeatable scoring—not another generic chat.

The product

What we are building

A self-contained workspace: Upload PDFs for a full audit pipeline (parse → embed → per-control LLM scoring), explore 1,242+ terms in a knowledge graph, learn via story tracks and flashcards, and refine Arabic text with glossary-aware support.

Platform — capabilities & modules

Below is what you can do today inside this application—each area opens from the tabs above or from the workspace.

Benefits

01

Evidence-backed audits

Scores are anchored to rubric controls with chunk retrieval—not generic summaries.

02

Runs on your stack

Local FastAPI + your NVIDIA API key; documents stay on your machine during processing.

03

Trilingual glossary

ICAIRE terms in Arabic, English, and French—aligned for policy and procurement language.

04

Learning loop

Graph, stories, and flashcards turn the glossary into structured study.

Services & modules

Loading glossary data...

Workspace

Statistics, picks from your glossary, and shortcuts to every tool.

Total Terms
0
Story Tracks
7
Graph edges
8,400
8 edge types
UML diagrams
1,180
auto-generated

Tools

Knowledge graph

Every term is a node; every edge is typed. Clusters auto-discovered via Louvain community detection. Node size = PageRank.

Story tracks

Chronological journeys through the glossary. Each track groups terms into narrated chapters for a cinematic learning experience.

Chapter 4

الترميز الموضعي

الآن لدينا سلسلة من المتجهات. ولكنّ المحول لا يعرف أيّها كان أولاً — فآليّة الانتباه تنظر إلى كل الرموز دفعةً واحدة.

هنا يأتي دور الترميز الموضعي: نُضيف إلى كل متجه بصمةً رياضيةً تخبره بموقعه في الجملة.

قد تكون هذه البصمة دوال جيبية أو متجهات مُتَعَلَّمة. النتيجة واحدة: الرمز الأول يختلف عن الرمز العاشر، حتى لو كانا نفس الكلمة.

Logic Architecture
Select a chapter to render its architecture.
Local JS Rendering · No external tokens used

Term detail

Everything the enrichment pipeline generated per term — metaphor, UML, prerequisites, graph relations, code, and math.

Loading term…

Flashcards & quizzes

Auto-generated distractors from the enrichment pipeline. Choose a difficulty level to focus your learning.

Which definition matches this term?
---
Card of -- streak 0 · next review soon