Case study · 02
Haqiqa
Agentic Arabic News Verification
Haqiqa verifies Arabic news by turning a claim into a structured evidence-search workflow. It searches for relevant sources, ranks the evidence, evaluates what the sources support, and returns a reasoned result with citations.
- Arabic NLP
- Agentic AI
- Evidence Retrieval
Problem
Arabic news verification needs more than a standalone true or false classifier. A useful result should show what evidence was found, which sources support or challenge the claim, and why the final result was produced.
Verification workflow
Haqiqa uses LangGraph to coordinate a five-stage agentic verification workflow. The process focuses on evidence and keeps framework details secondary to the result.
- Extract checkable claims
- Search the web for relevant sources
- Rank the collected evidence
- Evaluate the strongest evidence
- Return a final result with reasoning and citations
Evaluated ML foundation
Earlier model development used a dataset of 13,750 Arabic news articles. AraBERT reached 96.22% F1, 96.22% accuracy, and 99.57% AUC. XGBoost reached 94.50% F1, 94.51% accuracy, and 98.94% AUC.
These models belong to the earlier machine-learning layer and are not the full current verification workflow.
Product experience
The web app provides an Arabic-first RTL interface, verification results with evidence and citations, saved analysis history, and authentication. A Telegram interface offers an additional access point.
Verification quality depends on the evidence available online, so results should be read together with the cited sources rather than treated as an unquestionable authority.
Related areas
Arabic NLP · AI Evaluation · Intelligent Systems