Skip to content
B

Bio Bubble

Environment Care without Human care

Created on 10th December 2025

B

Bio Bubble

Environment Care without Human care

The problem Bio Bubble solves

The Problem It Solves

Indoor plants die easily because most people have no idea what’s happening inside the soil or the air around the plant.
Watering schedules are guesswork, humidity goes unchecked, temperature fluctuates, and people only notice problems after the plant is already damaged.
This system eliminates that guesswork by giving real-time visibility + predictive alerts.

What People Can Use It For:

  1. Prevent Plants From Dying

Most plants die from overwatering, underwatering, or humidity stress.
Your system tracks these conditions continuously and warns the user before the plant is harmed.

  1. Monitor Plant Health Remotely

People can check:

Temperature

Humidity

Soil moisture

Current stress

Future predicted stress

from anywhere via a cloud dashboard (ThingSpeak).

No need to manually check the plant every day.

  1. Get Predictive Notifications Instead of Guessing

Instead of reacting when a plant starts drooping, users get:

“Soil will dry out soon”

“Plant stress predicted to increase in 30 minutes”

It’s proactive plant care — not reactive.

  1. Helps Beginners Who Don’t Understand Plant Science

Most people don’t know ideal moisture/humidity ranges.
The system translates raw data into simple actions:

Move plant to shade

Give more light

Water soon

Stable condition

This removes the need for plant expertise.

  1. Reduces Maintenance Effort

No more:

Sticking fingers in soil

Overwatering

Checking humidity manually

Guessing light levels

The pod handles monitoring and decision logic.

  1. Makes Indoor Gardening Safer

For office buildings, elderly users, or students:

Prevent fungus from overwatering

Avoid dead plants that attract pests

Keep indoor air quality healthier through healthy plants

  1. Useful for Research & Education

Students can use the real-time data to study:

Plant stress behavior

Microclimate patterns

Soil moisture decay curves

Predictive AI models

It’s a mini environmental lab.

Challenges we ran into

Challenges I Ran Into

  1. Sensor Values Were Inconsistent and Unstable

The DHT11 and soil moisture sensor initially gave values that fluctuated heavily.
This made the AI model and stress detection unreliable.

How I solved it:

Added a pull-up resistor for the DHT11 data pin.

Averaged multiple sensor readings before sending to ThingSpeak.

Calibrated soil readings by comparing dry/normal/wet states manually.

This stabilized the data enough for meaningful analysis.

  1. ESP8266 Analog Pin Caused Wrong Soil Moisture Readings

The ESP8266 has only one analog pin (A0) and it accepts only 0–1V, but the soil sensor output is up to 3.3V.

How I solved it:
The module fortunately has an internal scaling resistor, so I tested readings in water, air, and soil to find the correct raw ranges.
Then I applied a custom mapping formula to derive reliable percentages.

  1. DHT11 Failed Randomly During Early Tests

Sometimes the serial monitor showed:
DHT read failed

Why it happened:

The sensor is slow

Needs correct timing

Requires a strong signal

Fix:

Increased delay between readings

Ensured solid wiring

Used the recommended DHT library with retries

Once corrected, readings became stable.

  1. ThingSpeak 404 / No Data Issue

My ESP8266 code sent data, but ThingSpeak didn’t update and kept returning “0” or “404”.

Cause:

Wrong API key

Wrong field order

Sending requests faster than 15 seconds

Fix:

Replaced incorrect key with the Write API key

Added a 20-second delay

Printed the final URL in Serial Monitor to debug

This immediately resolved the issue.

  1. Predictive AI Model Was Not Accurate at First

The initial model predicted stress poorly because the dataset was too small and sensors were noisy.

How I improved it:

Collected data continuously for hours

Cleaned outliers

Added features like “previous stress”, “hour of day”, and rate of change

Switched to Random Forest instead of simple regression

Prediction accuracy increased significantly.

  1. Vercel Deployment Gave 404 Errors

Frontend didn't load because the entry file was named App.html, not index.html.

Fix:

Renamed file to index.html

Completed Git commit properly

Pushed changes

Vercel redeployed automatically

After that, the dashboard loaded cleanly.

Discussion

Builders also viewed

See more projects on Devfolio