BMC recovers ₹13 crore with AI detection of irregularities in desilting

As the BMC spent a whopping ₹238 crore in 2025-26 on desilting across all nallahs -- minor, major and the Mithi, Bangar reasoned it makes sense to invest in AI so that the exercise would be foolproof

MUMBAI: Having Artificial Intelligence (AI) monitor vast swathes of photographic footage/evidence of pre-monsoon desilting, the Brihanmumbai Municipal Corporation (BMC) has identified multiple instances of violations that have saved the civic body over ₹13 crore. When the AI system detected violations indicating a possible irregularity in desilting, the BMC did not pay the contractor for the tainted trip.

Pre-monsoon drain cleaning work (Photo by Satish Bate/ Hindustan Times) (Hindustan Times)

Considering this foray into using AI as successful, particularly as monitoring desilting requires wading through legions of footage, additional municipal commissioner Abhijeet Bangar said the civic body will be investing more and bringing on board an agency for better AI detection for the next pre-monsoon desilting exercise.

“The most common violation the AI system flagged was when photos and videos of desilting were not uploaded, which made it an open and shut case,” said Bangar. “Another slight the system caught is when the time difference between an empty van and a van filled with silt was too little to be believable.”

The AI detection system also recognised desilting footage that was filmed from an old pre-recorded video from a different phone at the site to have correct GPS markers. This was spotted from the shadowing seen in the footage. In these cases, the BMC automatically deducted the payment for the desilting trip to the contractor, without seeking an explanation. Bangar said this was done to increase compliance from the contractors and ensure the job was carried out properly.

As the BMC spent a whopping ₹238 crore in 2025-26 on desilting across all nallahs -- minor, major and the Mithi, Bangar reasoned it makes sense to invest in AI so that the exercise would be foolproof. “For instance, AI can be trained to identify if there is a lot of dust in the process of desilting in the videos uploaded to flag debris being mixed. For this, it needs to be trained repeatedly and refined. An agency will help us perfect this,” he said.

Sources in the BMC also suggest that the stringent requirements of footage and AI checks have resulted in a backlog in the desilting target, particularly in the minor nallahs, with contractors shrinking the work due to the increased checks. Instead of being worried, officials said this would help identify how much desilting is required for the city next year to function in heavy rains.