They fit in the palm of your hand.
Every year, they fly from Mexico to Canada and back.
They are also listed as an endangered species in Canada.

You guess it right, it's the monarch butterflies.






The data was collected over the .... iNaturalist

The raw dataset contained 703,394 records spanning from the 1800s to 2025. The first thing I noticed was that more than half the records (371,805) had no recoverable date, meaning neither the eventDate field nor the year, month, and day fields were populated.

As visible in the chart below, data before 1980 is extremely sparse, so I restricted the analysis to 1980 onwards. The spike around 1999-2001 also stood out. Digging into it, it turned out to be a bulk submission from Monarch Watch, a dedicated monarch monitoring program based out of the University of Kansas that has tracked the species since the early 1990s. These records were aggregated annually and lacked day and month information, making them unusable for phenological analysis.
After removing records with unrecoverable dates, missing coordinates, duplicates, and pre-1980 observations, the dataset was reduced to 316,273 records. Given sparse coverage in the early years, the analysis was further restricted to 2010 onwards, yielding a final dataset of 312,235 records, roughly 44% of the original.

The maps below show monarch butterfly sightings across North America colored by month. The full migration corridor is immediately visible, with purple and blue dots cluster in Mexico during winter, greens push north through spring, and reds and oranges spread across the US and Canada through summer and fall. Comparing across periods, the density of sightings increases noticeably over time, largely reflecting the growth of iNaturalist as a platform rather than the actual monarch population. To better understand the migration pattern, I focused on the Atlantic coast corridor and traced the path monarchs take as they move north through Georgia, the Carolinas, Virginia, and up through New England.

The table below shows the number of sightings per state per year. The iNaturalist effect is immediately visible: records in every state jump dramatically after 2017, reflecting platform growth rather than monarch population changes.

Despite the uneven coverage, a clear pattern emerges when we look at early arrival dates. For each state, I calculated the 5th percentile sighting date across years with at least 100 records, effectively when monarchs first consistently show up. Only years with at least 5 valid years of data were included.

State	Avg Early Arrival	Valid Years
South Carolina	Mar 30	9
Georgia	Apr 17	7
North Carolina	May 07	9
Delaware	Jun 03	6
Virginia	Jun 11	10
Maryland	Jun 22	10
Connecticut	Jun 25	7
Massachusetts	Jun 25	9
New Jersey	Jun 26	9
New York	Jun 27	9
Maine	Jul 06	9

The results tell a clean story. South Carolina sees its first consistent sightings around March 30, Georgia around April 17, and the signal moves steadily northward through Virginia in June, reaching Maine by early July. The map below visualizes this progression along the corridor. This is the spring northward migration captured in citizen science data. While not perfect, it is remarkably coherent given the limitations of the dataset.




The Arduino collects the sensors' data and sends it to the backend (a Python FastAPI application) via asynchronous serial communication. The data is then processed and displayed in React on the frontend. I used a combination of WebSockets and HTTP for communication across the stack. I am fairly new to WebSockets, so it was really cool implementing a two-way communication channel between backend and frontend.

I used GPT-4o by OpenAI to handle text-to-speech (TTS) communication and give Lady Monstera her voice.


Seeing the plant’s internal state update in real time was absolutely thrilling! I tracked water and light changes over 5-second intervals, and let the plant comment on significant state changes. Here is how Lady Monstera recently handled an abrupt dimming of the lights:

And here is how she reacted to being watered! Notice how she records and updates the last time she was watered - just another way to remind you how sassy she is, and how she won’t forgive you for forgetting to water her!

• I opted for text-to-speech instead of realtime speech. Realtime was very sensitive to ambient noise, difficult to control and it quickly became too expensive to experiment with.
• It would be fun to gamify the experience, with multiple plants talking to each other, each with distinct personalities based on their needs. Maybe even connect hobbyist gardeners and let their plants gossip about their owners 🤭
• Ultimately, I hope this project helps us get closer to our houseplants and the nature around us, reduce houseplant neglect and make interacting with our green friends a little more entertaining and curiosity-driven.

The Waterloo Voice AI Hackathon was my first real-world hackathon and I loved every minute of it. Coding next to other hackers, exchanging ideas and simply absorbing the energy and enthusiasm in the room has left an indelible mark on me. It has given me the confidence to keep building, no matter how silly or small the initial idea might seem.

A big thank you to the organizers and judges, in particular Ian Pilon, Mike Bird, Ti Guo and Prabal Gupta for your dedication and support.

Keep hacking and see you all at the next one!