AI software stack processes defence and law enforcement data, analyses video, audio, and images, and generates intelligence from large unstructured datasets.
Defence, intelligence, and law enforcement agencies are dealing with a growing volume of unstructured data—from CCTV footage and audio recordings to images and telecom records—while existing forensic workflows remain largely manual, fragmented, and time-intensive.
An Indian startup Pi-Labs is building an AI-first software stack aimed at addressing this gap by accelerating how such data is analysed and converted into actionable intelligence. In a recent conversation with Ankush Tiwari, Founder and CEO of Pi-Labs, he said the company is developing an AI layer that sits over existing infrastructure to speed up forensic and intelligence workflows used by defence and law enforcement agencies.
The platform is designed to work with systems already deployed in the field, including surveillance cameras, microphones, and telecom data feeds, without requiring new hardware installations. Instead, it processes inputs from existing infrastructure to support faster analysis of operational data.
The product suite spans deepfake detection, video forensics, audio forensics, speech intelligence, data fusion, and a chain-of-custody system that tracks and secures digital evidence across agencies. According to the company, its deepfake detection system is already deployed in multiple law enforcement and inter-agency environments in India.
The core challenge the platform targets is scale, as investigators are required to analyse large, mixed datasets that are difficult to process using conventional tools. Ankush said, “The system is designed to reduce time spent on tasks such as scanning CCTV footage, matching audio samples, enhancing image quality, and identifying manipulated media.”
Deployments are built for air-gapped and on-premise environments, where cloud access is restricted due to security requirements. Hardware is provided through an OEM partnership with Dell, while Pi Labs builds the AI software layer on top. The company operates GPU-based infrastructure with more than 25 GPUs used for training and inference workloads.
Given the sensitivity of defence and forensic applications, explainability and auditability are central to the system. Outputs are designed to be traceable, with reasoning layers intended to reduce false positives in investigations. Ankush added, “Beyond model development, significant effort goes into inference optimization, deployment stability, and managing performance in disconnected environments.”
The company follows a defence-first development strategy, building for high-reliability use cases before extending into broader enterprise applications. It is also positioning its platform as part of a sovereign AI stack for India’s defence and intelligence ecosystem, with support for regional languages and dialects.
The startup said its long-term goal is to build a sovereign, India-first defence AI stack that can operate reliably in high-security environments with strict accuracy requirements.
