From plan. To pole. To nation.
MRAI is not a one-off science project. Round 2 took us to first runner-up on a breadboard; Round 3 is Prototype v2 — sealed, solar-powered and night-capable — on the way to a national roadside sensing network that costs less than a smartphone per pole.
Research, problem mapping, and system design.
- Identified blackspot zones from MoRTH 2024 accident reports.
- Selected Raspberry Pi 3B + ESP32 + LoRa SX1276 architecture.
- Defined three hazard classes: Rain, Obstruction/Proximity, Smoke/Gas.
- Sketched the first dashboard wireframe on paper.
Sensor integration, YOLO on Pi, and first dashboard.
- Wired HC-SR04, YL-83 rain, and MQ-2 smoke sensors on a breadboard.
- YOLOv8n TFLite running at 2–4 fps on Pi 3B, triggered by sensor gating.
- First ESP32 → Flask JSON bridge tested locally.
- Placed first runner-up in Round 2 with this build.
v2 was built, then shelved before bench testing finished.
- Built: soldered harness in an enclosure, no breadboard in the loop.
- Built: Pi 3B + ESP32 sensor bus, OLED status panel, LoRa SX1276 radio.
- Built: authenticated telemetry bridge posting into this dashboard.
- Paused: node disassembled due to lack of time and funding.
- Not completed: 72-hour soak run and staged hazard trials — no results claimed.
- Preserved: software platform, ingestion API, BOM and wiring docs.
Single-junction field test, once hardware is rebuilt and tested.
- One pole at a low-risk site with permission, line-of-sight coverage, and mains power.
- Real rain, dust, and smoke conditions streamed to this dashboard.
- Detect → alert latency measured end-to-end instead of estimated.
- Two weeks of recorded footage scored against ground truth.
Thermal vision, edge AI, and self-healing mesh.
- Add FLIR/Lepton thermal camera for 24-hour detection.
- Deploy Coral TPU / Pi AI HAT for sub-200ms on-device inference.
- Build self-healing LoRa mesh with automatic dead-node rerouting.
- Hardened solar + LiFePO4 kit rated for Indian summers.
City command dashboard and nationwide scaling.
- City command dashboard for traffic police and ambulance dispatch.
- GIS pole planner using accident-hotspot and line-of-sight maps.
- Open telemetry API for smart-city integration.
- NB-IoT / 5G fallback where cellular coverage exists.
What powers the next phases.
Every item below is a direct extension of the hardware stack we already have.
Low-cost FLIR module cuts through fog, glare, and complete darkness.
Nodes auto-reroute around a dead pole. No single point of failure.
20W panel + 12Ah battery keeps MRAI alive through 48-hour outages.
Adaptive sleep cycles extend uptime to 4+ days on battery alone.
Coral TPU replaces cloud inference for sub-200ms crash detection.
Unified web dashboard for police and ambulance with real-time heatmaps.
Accident-hotspot + line-of-sight maps recommend the next pole location.
Cellular backhaul for SOS alerts where coverage exists.
Documented JSON API so city engineers can plug MRAI into smart-city stacks.
A network that never sleeps, across every highway in India.
MRAI is designed to be cheap enough to deploy at scale, rugged enough to survive Indian summers, and smart enough to act before a human can call for help. That is the competition we are here to win.