KBTU and Digital Court Analytics: AI Supporting Judges

How a “smart” assistant helps judges quickly find similar cases and harmonize judicial practice

The national push toward digitalization and artificial intelligence, identified by the President of Kazakhstan as a key priority, led to the creation of a civil-case system called “Digital Analytics of Judicial Practice.” The system was developed by the Kazakh-British Technical University (KBTU) in collaboration with the Judicial Administration of the Republic of Kazakhstan and the National Chamber of Entrepreneurs “Atameken.” In essence, it is a “smart assistant” for judges. The program takes over the routine work of searching and analyzing large volumes of judicial acts, generating preliminary recommendations for each case. Although the law is the same for everyone in Kazakhstan, identical situations are sometimes interpreted differently across regions. This affects public trust in justice and overloads judges with manual analytics. The new system aims to solve both problems—reducing document-processing time and helping align judicial practice across similar cases. We spoke with KBTU developers and representatives of the Judicial Administration to learn how the system works and why it matters for the entire judicial system of Kazakhstan.

– In 2018, under a memorandum with the National Chamber of Entrepreneurs “Atameken”, the Department under the Supreme Court (now the Judicial Administration) tasked us with creating a system capable of “understanding” judicial texts, automatically identifying relevant decisions, analyzing court rulings, and generating automated analytical recommendations, – says KBTU senior lecturer Bobur Mukhsimbayev. At the first meetings, we discussed what exactly was required. One thing was clear: the program needed to “understand” the meaning of judicial texts and detect patterns, rather than simply search for keywords. Development took place in stages from 2018 to 2020. We worked iteratively: presenting prototypes, collecting feedback, improving the system. At the start, we immersed ourselves in the judicial system, studying business processes, decision-making logic, and the structure of judicial acts. The Supreme Court’s main objective was to achieve consistency in judicial practice: identical laws should lead to similar decisions in similar cases. The second objective was to reduce the burden on judges. Since 2016, the “Torelik” system and the Judicial Cabinet (which accepts claims electronically) had already been operating. This increased the flow of cases, and judges found it increasingly difficult to manually search for analogues among millions of documents.

– The developed system, “Digital Analytics of Judicial Practice”, became a recommendatory tool for judges. The system does not issue a verdict but helps navigate similar cases more quickly, provides semantic search, identifies anomalies, and uses mathematical models to predict possible outcomes—while the final decision is always made by the judge. For example, if a claim is filed over a flooded apartment, the judge describes the situation in legal language, and the program finds similar cases across the country—in Almaty, Shymkent, Astana, and other regions—showing what decisions were made by first instance, appellate, and cassation courts. This allows the judge to compare their own reasoning with established practice. To make this possible, we initially uploaded and anonymized around three million judicial acts. In just a few clicks, a judge can obtain a set of cases with the most relevant documents highlighted. This saves time and provides a factual foundation.

– Today the system is used for civil disputes—land matters, banking cases, and others where consistent judicial practice is especially important. Criminal and family cases are not yet included due to their specific nature and confidentiality requirements. I believe the system has already proven its usefulness: it makes the process more transparent and predictable. Of course, digitalization requires constant updating and adaptation, but the core idea—ensuring consistency and reducing routine work—remains essential.

– Our project consisted of several modules. My colleague Bobur was responsible for the part related to searching similar cases and detecting anomalies in judicial practice, – says KBTU senior lecturer Aibek Kuralbayev. – I was working on another important task — data anonymization. It was necessary to ensure confidentiality by removing personal data, names of individuals, company names, and other identifiers from judicial acts. I developed algorithms that detected such data in the text and properly masked it. My academic research is also focused on this area. The project team consisted of six people. Bobur and I mainly worked on building models and analyzing data — in other words, we were responsible for the system’s intellectual core. Other team members developed the interface, databases, document-upload mechanisms — everything that keeps the system running “under the hood.” In earlier years, when we still had access to the system, we saw from the logs that judges actively used it, and there were many different types of cases. Now we no longer have direct access to this analytical platform, but we are confident that the tool remains useful and in demand.

The Judicial Administration confirms that the “Digital Analytics of Judicial Practice” system is currently used for the analysis of civil cases and is regularly retrained on up-to-date judicial acts. According to the court, more than five million documents are included in the system; numerous analytical summaries have been prepared, and more than a thousand informational briefs have been compiled. They emphasize that the system’s purpose is to promote consistency in judicial practice and provide judges with an additional analytical instrument; each case, however, is still reviewed individually based on its circumstances. Regarding criminal cases, the court notes their specific nature: the process begins with the pre-trial investigation carried out by the police and the prosecutor’s office, while sanctions in the Criminal Code are imperative. Therefore, AI implementation should reasonably begin at the pre-trial stage within law enforcement agencies. Such a “smart assistant” does not replace the judge and does not eliminate individualized assessment. But it makes the process more predictable and clearer. Similar cases produce similar decisions, there is less manual routine, and more reliance on factual patterns. As a result, everyone benefits: judges save time, lawyers prepare their positions more accurately, and citizens better understand the logic of the system.

Today the Judicial Administration has gone further and, using the accumulated experience, is already independently developing other AI-based services.