AlHakikaNews

An AI-powered platform using a resource-efficient hybrid model to detect fake news in real-time, restoring trust in digital media across Africa.

Comment

Inspiration

Our inspiration was born from a frustrating reality: in today's digital world, a lie travels six times faster than the truth. This has created a crisis of confidence, where 70% of people feel confused about what news is factual, leading to social polarization and an erosion of trust in media. We saw this issue impacting our own communities in Morocco, a country that has become a key target for sophisticated disinformation campaigns. We were inspired to build a technological shield against this threat. We named our project AlHakikaNews—"AlHakika" being the Arabic word for "The Truth"—to ground our mission in a clear, unwavering purpose.

What it does

AlHakikaNews is a real-time platform designed to verify the authenticity of digital news content. It functions as a powerful tool for journalists, media houses, and the general public. Here’s what it does:

A user can submit any news article to the platform for an immediate authenticity assessment.

Our system performs a deep analysis, checking for plagiarism and semantic similarity against existing web content to ensure originality.

Using its core hybrid AI engine, the platform delivers a clear, real-time classification of the article as either "Real" or "Fake".

The ultimate goal is to provide a reliable, accessible, and fast fact-checking service that enhances journalistic integrity and creates a safer information environment.

How we built it

Our core technical challenge was to build a system that was both highly intelligent and highly efficient. Our solution was a Hybrid AI Architecture.

We knew a single approach would fail; a pure LLM system would be too costly and slow, while a classical ML system would lack nuance. Our hybrid model strategically combines:

Large Language Models (GPT & Gemini): For deep contextual and semantic understanding of the article's narrative and claims.

Classical Machine Learning (Gradient Boosting, Decision Trees): For rapid, low-cost, and efficient pattern recognition and classification.

This entire system is built on a modern tech stack, including Python and FastAPI for the backend, React and Next.js for a responsive frontend, and a Weaviate vector database for high-speed similarity searches. Crucially, the model was trained on our own custom-built, balanced dataset of over 44,000 annotated articles.

Challenges we ran into

Our biggest challenge was building within the spirit of resource-constrained computing.

Computational & Financial Costs: The primary hurdle was the prohibitive cost and latency of relying on external LLM APIs for analysis at scale. This financial constraint forced us to innovate on our architecture to ensure our solution would be affordable and accessible for African media outlets.

Data Scarcity: We initially struggled to find large, high-quality, and localized datasets for the African context. We overcame this by undertaking the difficult but necessary task of building our own comprehensive and balanced dataset from scratch.

Accomplishments that we're proud of

We successfully developed a functional MVP with a low-latency API capable of delivering results in approximately 500ms.

Our hybrid model achieved outstanding performance metrics, including 95.3% Accuracy and 95.9% Recall, proving its effectiveness.

Our proudest accomplishment is winning 3rd place in the prestigious 2025 DeepFake Challenge, organized by UM6P, Inwi, and StartGate. This provided incredible external validation of our technology and approach.

The success in the challenge has already generated significant interest, leading to productive follow-up meetings with the organizers to discuss concrete collaboration opportunities.

What we learned

Throughout this journey, our key learning has been that constraints drive innovation.

We learned that a hybrid AI model can outperform a single-minded approach by creating a pragmatic balance between state-of-the-art accuracy and real-world efficiency.

We learned that solving for resource constraints forced us to build a smarter, more efficient, and ultimately more scalable product.

Finally, we learned the immense value of external validation. Competing in challenges pushed us to refine our work and opened doors we couldn't have on our own.

What's next for AlHakikaNews

Our vision is to scale our impact from Morocco to the rest of Africa. Our next steps are clear:

Formalize Research Partnerships: We will actively pursue the research collaboration opportunity with UM6P's College of Computing to continue pushing the boundaries of detection models.

Integrate via Technology Partnerships: We plan to integrate our API with major media platforms and organizations like Inwi to protect users at scale.

Embrace Full Resource-Efficiency: Our long-term goal is to train our own models locally. This will drastically reduce our operational costs and our reliance on external APIs, making our solution truly sustainable and self-sufficient.

Expand Multilingual Support: We will expand our platform's language capabilities to serve the diverse media landscape across the continent.

Built With

Share this project:

Updates