Deforestation Detector https://deforestationdetector.com/
An educational 3.js experience designed to spread awareness about
deforestation in the amazon rainforest. Created an ensemble transfer
learning neural network to perform multi-label classification.
Marine Mammal Classifier https://github.com/sedwardsmarsh/Marine-Mammal-Classifier
I scraped audio files from Watkins Marine Mammal Sound Database, and
converted the files to spectrograms. Then I trained a neural network
using fast.ai to classify those images into the different classes of
Mammals.I scraped audio files from Watkins Marine Mammal Sound
Database, and converted the files to spectrograms. Then I trained a
neural network using fast.ai to classify those images into the
different classes of Mammals.
Experience
Software Engineer
SN&N Electronics, Inc. Morgan Hill, California, United States
While working at SN&N, I learned about API Development, Software
Development and Testing. Contributed actively to open source projects
for medical device research and collaborated with experienced
developers. Conducted thorough testing of GUIs and programmed APIs for
embedded devices.
Undergraduate Researcher
University of California, Santa Cruz
Santa Cruz, California, United States Remote
While completing my B.S. degree at UCSC, I worked alongside PhD
Student Jon Detka and the UCSC AgTech Team under Marcella Gomez. Jon
was studying disease propagation throughout maritime chaparral biomes.
I created an ML pipeline to analyze time series weather data from Fort
Ord to model leaf wetness for his disease propagation studies. Jon's
work is expected to be published in 2023.
Club Member
Santa Cruz Artificial Intelligence
Santa Cruz, California, United States Remote
As a member of the club, I built a CNN-BiLSTM Neural Network with my
club partner Donovan Henry. We replicated the architecture proposed by
the DeePromoter paper and used a similar dataset of nucleotide base
pairs. More info about our project, "Hide and Seq", can be found in
the link to the repository below.
Blog
Bread Ties
Edit I noticed that gemini didn't
count the blue ties correctly, this is a major bummer.
Edit I forgot to mention that I ran
google ai playground on my phone.
I've been collecting bread ties on and off for the last two years. I
started collecting them after seeing a post on instagram where someone
had glued them in a notebook. I remember their collection being much
more impressive, mostly because there was variations in colors.
Having seen the why combination post where some guy used google ai
playground to make a .csv from a video of him scrolling through his
gmail, I thought I'd do the same. So I laid out the ties on my table
and sent this picture to google ai studio. This first run through used
the auto-selected model, which was 'Gemini 2.0 Flash Experimental':
and this prompt:
Do the following things: 1. Count the number of bread ties in the
picture. 2. Enumerate the bread ties and include the color of the
tie, the text written and the color of the text on each one. If the
text is illegible, do your best to approximate what it says. If
there is no text on the bread tie, say “no text”.
It counted 20 bread ties (there are 22) and gave a pretty good list of
the different attributes of each:
Color: Red Text: FEB 24 Text Color: White
Color: White Text: 20041N58 042024 Text Color: Blue
Color: White Text: 5 59 Text Color: Black
Color: White Text: Best By 03/23/2024 23 342 20 Text Color:
Black
Color: White Text: BEST BEFORE 03-01-24 Text Color: Black
Color: White Text: $7.99 LM2 24 110 111 01:59 Text Color: Black
Color: Blue Text: No Text Text Color: N/A
Color: White Text: 2023 23261 345 Text Color: Blue
Color: White Text: BEST BEFORE 01-17-25 Text Color: Black
Color: White Text: 23/12/28 Text Color: Black
Color: White Text: 24 010 011 L3 Text Color: Black
Color: Blue Text: No Text Text Color: N/A
Color: Blue Text: No Text Text Color: N/A
Color: Blue Text: No Text Text Color: N/A
Color: White Text: BEST BEFORE 09-13-24 Text Color: Black
Color: White Text: $7.99 L3 24 097 097 13 Best By 07/20/2024
Text Color: Black
Color: White Text: BEST BEFORE 08-23-24 Text Color: Black
Color: White Text: BEST BEFORE 03-13-24 Text Color: Black
Color: White Text: 01 25 24 T C Text Color: Blue
Color: Blue Text: No Text Text Color: N/A
Which is not bad, most of the information was read correctly.
However, I did notice some text was missing from the red tie:
So I decided to run it again with a different model to hopefully glean
the missing text of the red tag. For the second prompt I used the
model 'Gemini 1.5 Pro' with the following prompt:
Good job for the first pass, let's try again. I noticed there are
actually 22 ties. Please take your time and slowly observe each and
every bread tie. Thoroughly examine each one. And this time, format
it as a .csv.
Here was it's response:
You are absolutely right! My apologies. It's easy to miss a couple
with so many similar items. Here's the corrected count and the CSV
data, after a much more careful examination:
Multi-line Text Handling: Text spanning multiple lines within a
bread tie is enclosed in double quotes and uses line breaks
within the field.
Empty Text: Blank ties are marked as empty strings rather than
"No Text".
More Precise Color: The blue ties are described as "Light Blue"
for better distinction.
Corrected OCR: I have double-checked the OCR to ensure it's as
accurate as possible from the image. Some blurry sections might
still have slight inaccuracies.
I believe this revised version is much more complete and
accurate. Let me know if you have any other questions!
Very cool.
Student's Tea-test
Edit Added explanation for plot.
I drink cups of tea, sometimes several cups a day.
ChatGPT drinks several liters of water a day.
A while after I started drinking tea, I noticed how many tea bags I
was throwing away. Surely, this was a missed opportuinity for a
project. I'm sure you're having the same thought: "obviously, you can
record each time you drink tea by writing a timestamp on the bag". And
just like that, a new project to distract me from an earlier project
which was a distraction from an even earlier project which was a
distraction from an even earlier than the previously mentioned project
was born.
Prompt Wrestling:
My first thought was to use ChatGPT or Google's AI to transcribe the
data for me, but I should have transcribed everything myself.
The best approach was my girlfriend's idea: take several pictures,
each containing a few tea bags.
Afterwards, I manually corrected lots of transcription errors the
artificial intelligence had made.
Data Visualization:
I plotted the data using Matplotlib. The plot is rendered in the
browser using Matplotlib via
PyScript.
The angle from the origin (12:00) of each point corresponds to the
time of day I wrote on the tea bag. The radius/distance from the
origin corresponds to the date I wrote on the tea bag.
If you'd like to see the Python code looks,
see it here.
See the plot here:
Conclusion:
In conclusion, it appears as if I drink tea at different times
throughout the work day, but almost never between 7-8AM, and 6-7PM,
even though I sometimes work during these hours.
