FarmSmart AI

An end‑to‑end IoT and ML platform for soil monitoring.

Hardware

• Raspberry Pi 4 with capacitive soil moisture sensor for live moisture readings

Data Storage

• Aggregates readings in a MongoDB cluster

Telemetry

• Sends moisture‑based recommendation messages to Azure IoT Hub

Machine Learning

• Trains a Random Forest regression on historical soil moisture data

AI Summaries

• Uses a Weather API and OpenAI GPT to generate plain‑language insights

Dashboard

• Flask app displaying live readings, regression charts, AI summaries, and Azure telemetry data

Startup

Install virtual environment

[py/python3] -m venv .venv

Activate virtual envrionment

source .venv/bin/activate

Install dependencies

pip install -r requirements.txt

Run application

[py/python3] app/app.py

NOTE: A little more info on the nature of a MetricsQueryResult response

Below is the hierarchal structure of a MetricsQueryResult response as well as a description for each object in the hierarchy. I had issues finding the documentation regarding what each of these objects meant, contained, etc.

Maybe I'm just a little stupid LOL.

• granularity: timedelta | None
  The time‑grain (window size) of each data point. For example, if you asked for 5‑minute granularity you’ll get one value per 5‑minute interval. Note that the service may adjust this grain up or down if your requested interval isn’t supported exactly.

• timespan: str
  The overall interval over which metrics were fetched, formatted as "<start_iso>/<end_iso>". This may differ slightly from your request if the service normalizes it.

• cost: int
  A relative “cost unit” indicating how expensive your query was. Higher numbers mean more work was done behind the scenes to satisfy the request.

• namespace: str | None
  The metric namespace queried (e.g. "Microsoft.Devices/IotHubs"). Namespaces group related metrics for a given resource provider.

• resource_region: str | None
  The Azure region where the target resource lives (e.g. "eastus2"). Helpful for cross‑region diagnostics.

• metrics: List[Metric]
  The actual metric results. Each entry corresponds to one metric name you asked for.

---

The Metric object

Each Metric in the metrics list has:

• id: str
  The full Azure resource‑ID for this specific metric definition.

• type: str
  The ARM resource type of the metric (usually "<namespace>/metricDefinitions").

• name: str
  The programmatic name of the metric (e.g. "d2c.telemetry.ingress.success").

• unit: str
  The unit of measure—common values include "Count", "Bytes", "Seconds", "Percent", etc.

• timeseries: List[TimeSeriesElement]
  One or more independent time series. If you split by dimensions (e.g. per device), you get multiple series; otherwise you usually get a single‑element list.

---

The TimeSeriesElement object

Each TimeSeriesElement represents one distinct series and contains:

• metadata_values: Dict[str, str]
  If you applied a filter or split by dimension, this dict maps each dimension name to its value (e.g. {"DeviceId": "abc123"}). If you didn’t filter or split, it’s an empty dict.

• data: List[MetricValue]
  A list of data points, one per time‑grain interval, each described by a MetricValue object.

---

The MetricValue object

Each entry in the data list has:

• timestamp: datetime
  When the data was sampled.

• average: float | None
  The mean over that interval (if you requested Average aggregation).

• count: float | None
  How many individual values contributed to the average (useful for understanding sparse data).

• minimum: float | None / maximum: float | None
  The min/max values seen in the interval.

• total: float | None
  The sum of all values (if you requested Total aggregation).

Not every field is populated on every call—you only get back the aggregations you asked for.