Shortlisted out of more than 400 teams at Smart India Hackathon 2025 for an AI and IoT rockfall early-warning system, architected end to end.
Sudden rockfalls in open-pit mines kill people, stop production and cost enormous amounts of money, and the industry response is overwhelmingly reactive — you find out when it has already happened. The brief was to make it predictive.
What was built is an early-warning system that estimates rockfall probability from multi-sensor data fusion at around 97% model accuracy — 98% AUC-ROC on the evaluation split — runs across an edge-and-cloud split so a mine keeps its warning when connectivity does not, raises real-time alerts, carries an SOS and emergency buzzer path for people on the ground, and reports into a live monitoring dashboard. No public dataset fitted the problem, so the training set was built from scratch inside a week.
The engineering here was end to end and single-handed: the front end, the backend, the database design, the SOS module, the buzzer integration and the web application experience.
It cleared the internal round and went through to detailed submission on the national portal, shortlisted from a field of more than four hundred teams. It did not reach the national finals — but the depth of what was standing at the end of it is a long way past a hackathon prototype, and it is a full case study on this site rather than a line on a list.
Divyakush Punjabi