Data Collector v3.0
BME68021.8 °C
Sensors connected
- Humidity64%
- Soil moisture38%
- CO₂812 ppm
Measures every minute
The Data Collector board sits under the seed tray and records seven variables — from temperature to soil moisture.
Plants don't talk.
Our sensors do.
Botanical Bytes makes growth measurable
Double opt-in, no spam. By submitting you agree to storage — privacy policy
Awarded by & featured in
AI that understands
the greenhouse
Botanical Bytes measures
around the clock
Temperature, humidity, CO₂, light, and soil moisture — every 60 seconds, to an SD card and the cloud.
Botanical Bytes measuring …
Writing data row #8,18221.8 °C
Botanical Bytes measures around the clock
Temperature, humidity, CO₂, light, and soil moisture — every 60 seconds, to an SD card and the cloud.
Cycle 12 running
Day 3 of 6 · next reading60 s
Keeps you
in the loop
Follow every cycle live and step in only where a reading falls outside the expected range.
Keeps you in the loop
Follow every cycle live and step in only where a reading falls outside the expected range.
And answers
your questions
How did cycle 12 compare? How much water was optimal? The dataset is public — just ask it.
Ask the dataset
And answers your questions
How did cycle 12 compare? How much water was optimal? The dataset is public — just ask it.
25%
more yield through the
optimal amount of water
The more cycles we measure,
the less we have to guess.
Open to everyone.
Rebuildable anywhere.
Board design, firmware, the neural network, and a complete sample dataset are all on GitHub.
„Junior Prize at Germany's national AI competition.“
„Finalists in Tübingen with the Plant Growth Optimizer.“
„Two special awards at the Jugend forscht state competition.“
How Botanical Bytes works
Sowing cycle 12
Cress seed10.0 g
Cycle comparable
- Data rows8,182
- Duration5.7 days
- Epoch940 / 1000
Learns from every cycle
Standardized sowings — always exactly 10 grams of seed — make cycles comparable. The neural network learns how they relate.
Experiment 12
Wateroptimized
Yield weighed
- Yield+25%
- Referencecycle 8
- Next cycleplanned
Finds the right amount of water
The data yields the optimal amount of water. The result: up to 25% more yield — and a system that gets smarter with every cycle.
Every seed counts. Literally.
Using OpenCV edge detection we analyze how seeds are spaced — spacing affects germination. The analysis is still done by hand.
The hardware in detail
Board design, firmware, network, and dataset — all open on GitHub.
Data Collector v3.0
Self-designed. Improved three times.
Our ESP32-S3-based Data Collector board measures temperature, air pressure, humidity, gas levels, CO₂, soil moisture, and brightness. The next generation adds pH and nutrient density — already designed, but it hadn't arrived by the 2024 deadline.
TemperatureBME680
Air pressureBME680
HumidityBME680
Gas levelsBME680
CO₂Channel A3
Soil moistureXH-4AK
BrightnessTEMT6000
pH · nutrient densityPlanned
3D model from the board design · 59.3 × 36.7 mm
FAQ
What is Botanical Bytes?
A student research project, started in 2023 for Germany's national AI competition (BWKI). We make plant growth measurable — with self-built sensor boards, standardized sowings, and a neural network.
Does the AI control irrigation yet?
No — not yet. Our current network predicts humidity from the other sensor channels. Real-time irrigation control is our stated goal and on the roadmap.
Where does the 25% more yield come from?
From experiments with the amount of water: we tested different amounts, weighed each harvest on a kitchen scale, and found the optimum. That was classic experimentation — not the neural network.
Why cress, of all plants?
Because it grows fast. One logged cycle takes about 5.7 days and yields 8,182 rows of minute-by-minute data across six recorded sensor channels. We also test radish microgreens.
What exactly does the board measure?
Temperature, air pressure, humidity, and gas levels (BME680), plus CO₂, soil moisture, and brightness. Every 60 seconds, to SD card and the cloud. The next board generation will add pH and nutrient density.
Is the project open source?
Yes. Board design, firmware, the neural network, and a full sample dataset are open on GitHub.
News from the
greenhouse.
Double opt-in, no spam. By submitting you agree to storage — privacy policy