Why does one cyber tip still take hours to process?
Kerala’s Counter Child Sexual Exploitation (CCSE) team receives around 4,000 cyber tips every month. Each one carries multiple image, audio and video files, and each one is worked largely by hand. Officers download and extract the files, review them one by one, flag sensitive content, search images and audio for clues that place the crime in Kerala, trace suspect IP addresses, and write to ISPs for subscriber details.
Three things make it costly:
- Manual bottlenecks. The officers are highly skilled but few, and the volume never lets up. Every tip waits its turn for the same manual steps.
- Human toll. Reviewing harmful material file after file is draining, and every minute spent on it by hand is a minute of exposure that could have been avoided.
- The wrong tools. Public chatbots aren’t built for sensitive case data, often refuse the questions investigators legitimately need to ask, and can’t capture the judgment experienced officers have built up over years in the field.
What if officers could hand their procedures to an agent?
The harness is an agentic orchestrator that lets non-technical officers automate their own work. It combines three ideas.
A chat interface officers already know
Officers ask questions in plain language — “find the ISP of this IP address”, “does this audio contain Malayalam?” — and the agent picks the right tool, runs it, and answers. In the demo, it identified an IP address as an Airtel connection in Kochi within seconds.
Tools any station can extend
The agent ships with investigative tools: IP geolocation and ISP lookup, audio language detection, image analysis and document-to-text conversion. Any police station or agency can add, create or modify tools to fit its own work.
Procedures that run themselves
This is the core of the system. Officers write a standard operating procedure as a short list of plain-English steps. The agent turns it into a workflow, runs it on uploaded case files, and reports live progress at every step. Officers can edit the steps at any time, and several procedures can run in parallel.
One upload, six steps, one report.
The first procedure automates CCSE tip processing end to end. An officer clicks Run, uploads the tip, and the agent works through six steps:
- Open and fingerprint the case.
- Check whether the photos originate from Kerala.
- Check whether the audio contains Malayalam speech.
- Extract suspect IP addresses and timestamps from the report, and trace each to a location and ISP.
- Draft emails to the ISPs requesting subscriber details.
- Clean up and file a report to a central database.
The agent handles the detail itself. To find the IP addresses, it converted the report from PDF to text without being asked, then chained several lookup tools together. The officer sees progress against each step of their procedure, not the technical noise underneath. In the demo case, both the audio and the images pointed to a Kerala connection, and two IP addresses traced to Kochi.
A case dashboard gives supervisors a bird’s-eye view of every tip: what’s in progress or complete, the platform it came from, the ISPs involved, and a severity rating. Officers define the severity logic themselves in plain English, and any selection exports to a spreadsheet.
From working every file to deciding on the evidence.
The harness moves officers from doing the legwork to reviewing what the agent found.
Time back for investigation
Hours of downloading, scanning and lookups per tip become a single run. Officers spend their time on the judgment calls only they can make, and less of it looking at harmful material by hand.
Knowledge that scales
A written procedure captures how an experienced officer works. Once it exists, any station can run it, so best practice no longer lives in one person’s head.
Humans stay in charge
Every run is broken into named steps with visible progress and a written report, so officers can see how each finding was reached. At HacKP, Bjørn R. Watne (Global CISO, INTERPOL) and Karolina Mason (Specialized Officer in Safeguarding, INTERPOL) pressed on how to balance human judgment and AI in child safety. The design answer: the agent surfaces evidence; officers decide.
Built together at HacKP 2026.
HacKP is the Kerala Police Cyberdome’s hackathon. This year, more than 50 participants and dozens of mentors came together to build the harness in 72 hours. Neuralcraft was proud to serve as the official AI Partner, with Niyas Mohammed leading the product.
The team demoed the system to senior leaders in Kerala law enforcement and received a memento from Shri Ravada Azad Chandrasekhar IPS (State Police Chief, Kerala), Shri Manoj Abraham IPS (DGP and Director, Vigilance and Anti-Corruption Bureau) and Shri Prakash P IPS (IGP, SCRB). Early deployments are now being planned.
The six years mattered more than the 72 hours. Agentic AI only earns a place in police work when it’s built around how officers actually work — and when the officer, not the agent, makes the call.