In the shadows of a bustling city, a skilled pickpocket swiftly retrieves a wallet from an unsuspecting individual’s pocket in a dimly lit alley. The victim, unaware of the theft until later, rushes to the nearest police station, hoping to recover the wallet and its valuable contents.
This scenario echoes throughout India, where law enforcement struggles amid a deluge of unstructured data and outdated record-keeping methods.
Enter a Gurugram-based tech startup poised to revolutionize law enforcement in India. Staqu Technologies, established in 2015 by a team of AI experts, has been quietly pioneering change. Their flagship product, Trinetra 2.0, integrates a potent language model named Crime GPT, set to transform how security forces tackle criminal investigations.
“Atul Rai, Co-founder and CEO of Staqu, shared, “When we consider big data, videos and images dominate the internet landscape. Our aim was to harness India’s vast data and elevate the intelligence of analog cameras.”
The Genesis of Crime GPT
Rai, an AI researcher with a degree from the University of Manchester, teamed up with Anurag Saini and Pankaj Sharma to confront the unique challenges facing Indian law enforcement.
In 2018, the company introduced Trinetra 1.0, enabling police forces to digitize criminal records, including crucial photographs and details. This laid the foundation for the next phase: leveraging AI to unlock the data’s potential. “Trinetra 2.0 expands upon its predecessor by integrating GPT to analyze unstructured text data,” explained Rai.
Rai elaborated that manually sifting through thousands of physical documents proved cumbersome and time-consuming. Enter Crime GPT. Rai’s team developed a customized large language model trained on an extensive database of over 900,000 criminal records from Uttar Pradesh. By incorporating Retrieval-Augmented Generation (RAG) techniques, the model delivers precise, factual responses.
Explaining the workings of Crime GPT, Rai noted its ability to interpret natural language queries and extract relevant information from unstructured data, such as First Information Reports (FIRs) and interrogation records.
For instance, if a police officer seeks a 21-year-old suspect facing charges under Indian Penal Code (IPC) Section 307 (attempt to murder), Crime GPT navigates the database, retrieves pertinent records, and furnishes comprehensive details of the suspect’s history, associated offenses, and even lists police stations where offenses are recorded.
Rai emphasized Crime GPT’s role as an invaluable assistant, particularly for police forces lacking digital platforms. “Today, police have a virtual assistant,” he remarked.
The Practical Challenges and Solutions
Staqu confronted practical challenges in deploying such technology in India, where data primarily exists in regional languages and unstructured formats. Rai highlighted the hurdle of Hindi-language data, addressed by developing AI models handling tasks like optical character recognition (OCR), translation, and tokenization, ensuring seamless processing of diverse police data inputs.
Ethical considerations remained paramount, with a dedicated team from the Uttar Pradesh Special Task Force (UPSTF) monitoring system outputs to mitigate bias issues. Rai stressed the importance of training AI models on filtered data for ethical AI engineering.
The Impact and Future Outlook
Staqu’s solutions have already made waves across India, with nine state police forces actively utilizing the Trinetra platform. The positive response has spurred demand from additional states, including Madhya Pradesh, Chhattisgarh, and Tamil Nadu.
Additional Director General of Police (Law and Order) at UP Special Task Force, Amitabh Yash, hailed the integration of a digitized criminal database with AI-powered Crime GPT, lauding its role in enhancing investigative capabilities and operational efficiency.
Looking ahead, Rai envisions audio and video analytics as Staqu’s future priorities. He envisions a future where every conventional CCTV camera evolves into an intelligent, AI-powered device, generating real-time insights and alerts.