Helping a Ukrainian NGO Combat Disinformation

Executive Summary 

Goals: Collect data on open-source russian media known for promoting state-aligned messaging, aggregate and categorize the war crimes evidence.

Solution: Ideation, design, and deployment of a disinformation analytics web portal with quick search and convenient interface for media monitoring. 

Benefits: The NGO enhanced its services helping Ukrainian investigative authorities with fine-grained data on open source russian media spreading disinformation and disinformation. 

Services delivered: Design, implementation, and maintenance of a system to monitor media outlets spreading biassed or manipulative content, and to identify presenters associated with disinformation and ideological messaging. 

About the Client 

NGO “Change Together” was founded in 2016 by an initiative group of employees from the Ukrainian STB TV channel. The organization focuses on media-related projects and humanitarian initiatives, with a strong emphasis on defending human rights. Its core mission is to protect and promote fundamental human rights, particularly in the context of combating disinformation and safeguarding the digital and informational rights of Ukrainians. Through its work, the organization plays a vital role in strengthening media literacy, fostering informed public discourse, and supporting Ukraine’s resilience in the information space. 

• Client: NGO “Change Together” 
• Location: Ukraine 
• Industry: NGO 
• Website: changetogether.org.ua 

Business Challenge  

The owner and propagandists of Radio-Télévision Libre des Mille Collines (RTLM) were sentenced in The Hague only after 25 years of the Rwandan genocide due to the complexity of the evidence collection process. Nowadays, russian disinformation is way more sophisticated and has many media streams that the state has been using to strengthen its narratives for over a decade.  

Our experts from Tietoevry Create teamed up with “Change Together,” a local non-profit organization, to create a solution that would collect and process existing disinformation evidence automatically in order to offer researchers, prosecutors, lawyers, and judges refined well-structured data on the violations of international humanitarian law committed before and during the war on the sovereign territory of Ukraine.  

One of the key challenges was to design a system capable of processing a large volume of video files simultaneously (up to 200 files at once). To meet this requirement, the system architecture was carefully tailored to ensure high performance with minimal latency. It was also built with horizontal scalability in mind, allowing for seamless expansion if the budget allows.  

Another critical requirement was implementing ultra-fast search. The system had to deliver search results within milliseconds, ensuring users could quickly access relevant data across a vast data set. We managed to tackle the challenges at hand mainly because of prior experience working on similar systems. E.g., we developed an IoT platform for a large Ukrainian agriculture business, helping our client to monitor thousands of machineries working in the fields in real time. 

Solution & Business Value 

Having extensive experience in building AI-driven products and systems, we understood from the outset that selecting the proper models would be critical for the project’s success, especially in terms of cost-efficiency. Even basic operations like voice-to-text or translation involve a vast array of tools and models, all with varying capabilities and pricing points. 

Before kicking off development, our team invested the first 3 months conducting extensive research and market analysis to identify the most suitable and cost-effective solutions. Choosing the wrong tools at this early stage could have led to significant budget waste and technical setbacks. 

Through this research, we adopted a hybrid approach, combining Microsoft solutions with other third-party tools to optimize both performance and costs. This foundational phase was pivotal to the project, requiring our data science team to rigorously evaluate and test multiple models to ensure the right fit. 

Thanks to this thorough preparation, we were able to deliver the final product, the Words of War platform, in just 4 months, keeping the total project cost at a highly competitive level. This initiative now stands as a flagship example of how careful planning, technical due diligence, and smart resource selection can lead to efficient, scalable AI solutions, especially for projects dealing with large-scale data processing.  

Our team had a number of benefits due to the courtesy and support of Microsoft Corporation, which offered its Azure cloud capabilities such as: 

• High-quality Cognitive Search 
• High-quality Cognitive Services for Text 
• Convenient and effective MLOps platform (Azure ML) 
• High-quality and performance of OpenAI 
• Availability of ChatGPT 4.0 for 32k tokens 
• Availability of high amount of GPU processing resources (120+ of Nvidia T4 GPU VMs) 
• Unlimited size of Cosmos DB 

In addition, Microsoft provided legal and business expertise covering Responsible AI principles and approaches, technical consulting support from in-house architects, and on-demand further support of hosting and system development. 

As a result of the collaboration, now this system is likely the most extensive of its kind focused on russian disinformation, containing open-source media data dating back to 2012, with a part of the system hosted in the public domain. It serves as a critical tool for the Office of the Attorney General of Ukraine and both Ukrainian and international media. Work is ongoing to promote the adoption of the tool by the other Ukrainian governmental bodies. 

Originally engineered with the defense sector in mind, Tietoevry Create’s system is highly adaptable and can also support the needs of customs authorities, border control, and other national security agencies. Its robust architecture and long-term data coverage make it a valuable asset for identifying, analyzing, and responding to disinformation and manipulation of public opinion. 

Technical Details  

A 7-month pro bono endeavor for the NGO “Change Together” had our experts working on a tailored system that has extensive functionality, covering: 

• Voice-to-text transcription (technology under NDA) 
• Speaker diarization / detection (technology under NDA) 
• Speaker identification (technology under NDA) 
• Automated transcription QA (technology under NDA) 
• Key phrase detection and NER (technology under NDA) 
• Signs of crime detection and classification (technology under NDA) 
• Narrative detection and classification (technology under NDA) 
• Search (technology under NDA) 
• Translation to English (technology under NDA) 
• Detecting text similarities, trends and dependencies (technology under NDA) 

We orchestrated the project with the help of 

• PM / Solution architect 
• 3 data scientists 
• 1 front-end & 2 back-end developers 
• UX / UI designer 

The engineers and the designer had a lot of functionality to build on the fly and managed to release the final product with zero necessity for quality assurance. 

Technologies & Tools 

Due to the sensitive nature of the Project, please reach out to our experts for a precise description of the technological stack we used.