Haqiqat AI

Evidences to recheck
how it looks (FACT CHECKER)
Urdu feature
Misinformation Detection and AI Score

Inspiration

Misinformation spreads 6x faster than truth on social media. In Pakistan and across the Global South, fact-checkers are few, overworked, and under-resourced. Manually verifying a single claim can take hours. We built Haqiqat AI because truth should not be a luxury that only well-funded newsrooms can afford. The name itself says it all: Haqiqat means reality in Urdu, and we believe everyone deserves access to it.

What It Does

Haqiqat AI is an agentic fact-checking system that extracts claims from any text, article, or statement automatically, searches trusted sources across the web in real time, verifies each claim using LLMs with transparent reasoning, and returns a verdict of True, False, Misleading, or Unverifiable with cited evidence. It scales to thousands of claims daily at 480x the speed of manual fact-checking.

How We Built It

We built Haqiqat AI using an agentic multi-step pipeline. LangGraph handles the orchestration of the verification workflow as a stateful agent graph. LLMs via Groq and LLaMA power claim extraction, reasoning, and verdict generation. Web search tools retrieve real-time evidence from authoritative sources, and n8n manages workflow automation and pipeline orchestration. Each claim flows through a sequence of Extract, Search, Reason, Verify, and Output, fully automated and fully auditable.

Challenges We Ran Into

The biggest technical challenge was hallucination control, as LLMs would sometimes generate confident verdicts without proper grounding in retrieved evidence. Source reliability was another obstacle, requiring us to filter low-quality or biased results during web search. Claim decomposition proved complex when breaking multi-part statements into verifiable atomic units. We also had to carefully balance latency against accuracy to keep the pipeline fast without sacrificing reasoning depth. Extending the system to support Urdu and other local languages added another layer of complexity.

Accomplishments That We Are Proud Of

We built a fully agentic fact-checking pipeline from scratch and achieved a 480x speed improvement over manual verification workflows. Every verdict the system produces is explainable, with transparent reasoning and cited sources rather than just a label. The architecture is scalable enough to process thousands of claims daily, and we laid the groundwork for multilingual fact-checking in underserved language markets.

What We Learned

Agentic AI is only as good as its orchestration logic. Prompt design and graph structure matter enormously. Fact-checking is not purely a retrieval problem; it requires structured reasoning chains built on top of retrieved evidence. Real-world misinformation is messy, and building for edge cases such as satire, opinion, and partial truths is where the most important work lives. We also learned that speed and explainability are not opposites. With the right architecture, you can achieve both.

What's Next for Haqiqat AI

The next phase includes a browser extension that lets users verify claims inline while browsing, full Urdu and regional language support to serve Pakistan's majority vernacular population, and a media API that allows newsrooms to integrate Haqiqat AI directly into their content management systems. We are also pursuing partnerships with fact-checking organizations and government transparency bodies, and plan to launch a misinformation trend dashboard that tracks what false narratives are spreading and where.