Legal AI Production RAG

Sanad
Arabic Legal QA System

A production-grade Arabic legal AI assistant powered by hybrid retrieval-augmented generation (RAG). Delivers verifiable, citation-backed answers grounded in verified civil, criminal, and labor codes.

49.4RPS

Throughput

Locust load-tested

418ms

Latency (P95)

Fast hybrid pipeline

0.0%

Error Rate

14,820 requests

0.90

Faithfulness

Gate threshold 0.75

1,149

Civil Articles

Egyptian Civil Code

78%

Test Coverage

64 pytest suites

Key Capabilities

Engineered for high factual fidelity, legal verification, and sub-second latency

Exact Statutory Citations

Every generated answer links directly to verified law articles, statutory numbers, and official codes.

Hybrid Retrieval (Dense + BM25)

Combines semantic dense embeddings with sparse BM25 keyword matching for rigorous legal term fidelity.

Anti-Hallucination Guardrails

Strict confidence thresholds force the system to decline speculation if no authoritative statute is retrieved.

Legal Terminology Expansion

Specialized Arabic legal thesaurus that maps layman queries into precise juridical and statutory jargon.

Legal Corpora & Coverage

Curated, structured indices parsed and embedded with dense chunking strategies

Egyptian Civil Code

1149+ Articles

Penal & Criminal Procedure Code

500+ Articles

Labor & Employment Laws

Official Statutes

Court of Cassation Rulings

Indexed Precedents

Tech Stack

Python

Backend Engine

FastAPI

API & Streaming

Hybrid RAG

Dense + BM25 Search

Legal LLM

Legal Reasoning

Vector DB

Indexed Law Articles

Guardrails

Attribution & Accuracy

Disclaimer: This platform is developed as a research and decision-support assistant for legal professionals, and does not constitute formal legal counsel.

Explore Sanad Legal QA

Check out the code, chunking methodology, and hybrid RAG implementation for Arabic statutory retrieval.

View Repository on GitHub