The Uttar Pradesh Police has introduced an artificial intelligence–based policing application, YAKSH, as part of its efforts to modernise crime prevention, investigation and beat-level operations across the state. The platform was inaugurated by Yogi Adityanath and unveiled during the two-day senior police officers’ conference Police Manthan 2025.
The closed-door launch, held as a government event, reflects the strategic and sensitive nature of the system, which is expected to play a role in law enforcement decision-making at scale. Officials view the deployment as a step towards more data-driven policing as India’s law enforcement agencies explore the use of AI in public safety.
Multimodal AI at the core of YAKSH
YAKSH is built on JARVIS One, a multimodal artificial intelligence platform developed by Staqu. The platform is designed to process video, audio and text data simultaneously within a unified system, allowing law enforcement personnel to analyse multiple evidence formats together rather than in isolation.
According to information shared around the deployment, JARVIS One operates entirely on internal police datasets, without drawing on external data sources. This design choice is intended to address concerns around data sovereignty, security and regulatory compliance—key considerations as AI adoption expands within government systems.
Using the platform, officers can search FIRs and case documents using natural language queries, conduct facial recognition from photographs, match voice samples for investigative leads, and analyse CCTV footage, audio recordings and textual records in combination. The system can also generate AI-led insights based solely on verified police data.
This marks an evolution from earlier JARVIS deployments, which were primarily focused on video analytics. The current implementation extends AI support across crime investigation, public safety and operational policing.
Verified data creation at the beat level
One of the distinguishing aspects of YAKSH is its approach to data validation. Criminal records are created at the police station level and are physically verified by beat officers before being incorporated into the AI system.
When an accused individual is registered in one district and linked to another, YAKSH generates alerts for the relevant district and beat officer, triggering on-ground verification at the person’s residence. This process is intended to ensure that AI models built on the platform rely on continuously validated, real-world data rather than static or unverified records.
Officials involved in the project view this as a way to improve the accuracy and reliability of AI-assisted policing, particularly in a large and diverse state like Uttar Pradesh.
AI-driven criminal ranking and network analysis
YAKSH also introduces an AI-based criminal ranking mechanism. Offenders are categorised and colour-coded based on factors such as crime severity, repeat offences, use of weapons and behavioural risk indicators. These rankings are updated dynamically as new, verified data is added to the system.
In addition, the platform supports gang and network analysis, automatic linking of FIRs and associates, and real-time alerts related to jurisdiction changes or movement of suspects. The system can also identify high-priority offenders at police station, district and state levels, helping authorities focus resources on repeat and high-risk individuals.
CrimeGPT brings generative AI to field policing
At the operational level, YAKSH includes a generative AI interface known as CrimeGPT. The tool allows officers to query crime data using everyday language, reducing reliance on complex databases or technical commands.
Investigators can search for criminals, patterns, gangs or historical incidents through a conversational interface, making advanced analytics more accessible at the field and beat level. This is particularly relevant in a policing environment where digital skills and access to specialised tools can vary widely.
Adoption beyond Uttar Pradesh
The underlying JARVIS One framework is also being adopted by India’s Cyber Crime Coordination Centre (Indian Cyber Crime Coordination Centre) for cybercrime intelligence and investigation. In cybercrime cases, where audio, video and textual evidence often intersect, multimodal AI capabilities are seen as increasingly relevant.
This broader adoption suggests that the technology is being evaluated for use beyond physical policing, as law enforcement agencies explore AI’s role in addressing both conventional crime and emerging digital threats.
A cautious step toward AI-led policing
While YAKSH represents a significant step in AI adoption for policing, its design choices—particularly around internal data use and human verification—indicate a cautious approach. Rather than fully autonomous decision-making, the system positions AI as an assistive layer, augmenting human judgement with faster analysis and broader visibility.
As more Indian states experiment with AI-driven governance systems, YAKSH is likely to be closely watched as a case study in how advanced analytics can be integrated into law enforcement while balancing accuracy, accountability and trust.