Overview
ForestNetra covers the whole pipeline: firmware on the sensor nodes, a backend that stores and analyses the readings, and a dashboard that shows decisions together with the factors behind them. I am building it as a Research Intern at IIT Kharagpur, and I am exploring TinyML and edge inference for on-device decision-making.
How it works
1. Sense
ESP32 nodes read soil, aerial and weather sensors.
2. Transmit
LoRa (SX1262) and RS485 carry long-range, low-power telemetry from the field.
3. Ingest
A FastAPI and SQLModel service stores every reading.
4. Infer
LSTM and PatchTST models make the decision, and Qwen 2.5 explains it in plain language.
5. Act
A Next.js dashboard shows each decision with the factors behind it.
Backend
FastAPI, SQLModel, PyTorch and Ollama. It exposes ingestion and history endpoints, an ML inference layer, and local LLM advice with explainable (XAI) factor tags, so a recommendation is never a black box.
Dashboard
Built with Next.js 16, React 19 and TypeScript: a live node grid, critical alerts, action history, CSV export, and a responsive layout from desktop to phone.
Tech stack
ForestNetra is ongoing research at IIT Kharagpur.