FactMata

As a software development company, we are aware of the responsibility that comes with developing solutions that can affect how information is shared and consumed online. When FactMata approached Geniusee to help optimize the AI system that detects misleading claims, our team saw an opportunity to positively influence the spread of accurate information.

By implementing our expertise in machine learning, natural language processing, and data analysis, we worked closely with FactMata to create web scrapers and optimize data pipelines and the project’s overall performance.

Services:

Industry:

AI Technology

Location:

United Kingdom

Dhruv Ghulati

CEO at Factmata

Business context

FactMata, a startup aimed to build AI systems that uphold principles of transparency and accountability, had a great mission — their services analyzed and identified likely unreliable information on the internet. Well-known investors like Mark Cuban, Biz Stone, and Mark Pincus recognized such potential.

In November 2022, Factmata was acquired by Cision — one of the biggest companies in the field of public relations services.

In our cooperation with Factmata, Geniusee delivered top-grade tech solutions to help this project scale by continually improving data pipelines and algorithms. AI systems were trained on massive datasets to detect signals correlated with false or misleading claims.

The mission behind our work was to apply AI responsibly and deliver real value to the digital information ecosystem.

Key challenges:

Obtaining large datasets to train the models.
Processing data from different sources of information.
Developing a user-friendly online dashboard.
Addressing issues related to data quality.
Data enrichment with third-party data sources.

Work approach

Scrum

Sprints, timed work periods, typically lasted two to four weeks. At the start of each sprint, we held planning meetings where teams committed to completing a defined set of tasks. Teams met regularly to review progress, address any blockers, and make adjustments as needed to achieve the sprint goal.
Integration with unstructured data process

We aggregated data from various public data sources, including news articles, social media posts, and forum discussions. This was done to build comprehensive datasets for training FactMata’s machine-learning models.
Complex data processing

In this part of our work, we applied an iterative approach. Our team ingested data from various sources and then filtered it for relevance and accuracy. We used algorithms to detect patterns and insights. After that, we evaluated outputs and refined algorithms to improve precision. This process was repeated until the models achieved optimal performance.
Performance optimization and scalability

We tuned FactMata’s database to enhance query performance and scalability. Our team optimized source code by reducing redundant logic, caching frequently used values, and improving algorithms. That way, we improved the platform’s speed and efficiency.

We aimed to create and refine a user-centered design that met key business objectives and end-user needs. Our constant collaboration, feedback, and revisions were instrumental to the success.
We worked on constructing robust data pipelines to ingest, clean, and annotate the massive datasets required to train AI models. These datasets contain examples of statements, claims, and other information from across the internet.
The front-end development process involved building the user interface to allow users access to Factmata’s solutions. The UI provides an easy way for people to upload content to be analyzed, view analysis results, and receive recommendations for improving or removing potentially misleading information.
We developed scalable solutions so they could handle up to 3 million records for one company. In particular, it was important to distribute the components to different parts of the cluster to make it work efficiently.
To develop FactMata’s data engineering capabilities, our team focused on building a robust data pipeline and storage infrastructure. We aggregated data from a variety of public data sources and preprocessed it. We optimized our resources and developed solutions with Kubeflow. As part of our data engineering operations, we also migrated processes to new data storage.
We maintained close contact with FactMata through sprint planning sessions and sprint reviews. This allowed us to gain a deep understanding of requirements and make any necessary adjustments to our development process

Project tech stack

Tools:

AWS

Geniusee team

Product team:

Development team:

Are you looking for full cycle product development?

You found what you were looking for. Just fill out the contact us form.

Features

Analysis of content quality
FactMata’s algorithms analyzed the accuracy of claims and statements across the internet. By cross-referencing claims with a vast database of factual information from reputable sources, FactMata could automatically detect and flag potentially false information.
Narrative monitoring
Through this feature, we can group similar opinions to form narratives, which one can view to learn what is being said internationally about a topic or a brand. It helped identify and understand the individual opinions being shared on online platforms.
User-friendly online dashboard with detailed results
The dashboard included analytics on moderation decisions, fact-checking accuracy, and user management tools. Additionally, it offered search functionalities and integration options, ensuring efficient content moderation and fact-checking processes.

Results

Our team has improved scalability, so we were able to receive more data on a constant basis.
We reduced the cost per record by switching to a serverless architecture.
We have created a large number of integrations with third-party content platforms.
Our experts also improved AI technology performance due to the decision to switch to NLP transformers.