Most hackathon projects die the moment the demo ends — a flashy prototype held together with tape, abandoned by Monday. I wanted the opposite: to take a hard, real problem and build something that could actually stand up as a system. That's how my AI + IoT rockfall prediction system got shortlisted at Smart India Hackathon 2025.
The problem worth building for
Rockfalls kill, and the tragedy is that they're often predictable — the slope gives warning signs before it fails, if you can read them. The catch is that no single sensor reads them reliably. Displacement, vibration, moisture: each is noisy, each cries wolf, and a system that false-alarms gets switched off right before the one time it's real.
That's a genuinely hard problem, not a weekend toy — which is exactly why it was worth attacking properly instead of faking a demo.
Building it like it mattered
The system is a multi-sensor data-fusion pipeline feeding a real-time hazard model. The core insight is fusion: no sensor is trusted alone, but weighed together their noise starts to cancel and a real pattern emerges — this combination, trending this way, means evacuate now. On the held-out evaluation it reached 98% AUC-ROC.
Two decisions kept it from being a toy:
- Fusion over thresholds. A trip-wire on one sensor is a demo; a model reasoning over the sensors as a system is a warning you can act on.
- Edge-first, because latency is safety. Inference runs close to the sensors, so a warning doesn't depend on connectivity that fails exactly when you need it — on a remote slope, that's the difference between a real system and a lab result.
What the hackathon taught me
The best thing a hackathon can do isn't teach you to build fast — it's force you to pick a problem that actually matters and build something honest against it under pressure. Getting shortlisted at SIH 2025 mattered less than proving to myself I could take a life-safety problem and produce a system with real engineering behind it, not a facade.
That instinct — build course projects and hackathon projects like they'll ship — is how a weekend idea becomes something you're proud to put your name on. The full pipeline is on the project page.
👉 See the system: www.divyakush.com/projects/rockfall-prediction
Divyakush Punjabi — Full-Stack & AI Systems Engineer
🌐 https://www.divyakush.com · 💼 LinkedIn · 💻 GitHub
Top comments (0)