Electrical Engineer & AI Enthusiast

I am passionate about all things circuit design and how AI is shaping our future

Contact Me

Giant LED Hannukah Display

Made up of a combined ~5000 individually addressable neopixel RGB LEDs.

Powered by 4 Arduino compatible Teensy microprocessors.

5 power supplies delivering a combined 65 Amps at 5 Volts.

~2000 lines of custom C code.

Almost 200 square feet of displays.

Wiring box for our newest project.

AI Handwritten Number Detector

Number Detector GUI

  1. Handwritten numbers are placed under the camera.

  2. The Capture Number button is pressed.

  3. The LLM detects writing and predicts the numbers.

  4. The Clear button resets the system.

  • Threshold Value sliding bar can be used to fine tune the system if its prediction is incorrect.

Number Data Processing

To properly detect the number, data processing to standardize the format must be performed.

The data processing was performed in 6 steps:

  1. The image is converted to grayscale. This simplifies the degrees of similarity needed for detection from 3 (RGB) to 1.

  2. A Gaussian Filter is applied to smooth out the data.

  3. Thresholding is applied to split the values into a binary 0 or 255 for simplified scaling in step 5.

  4. The colors are inverted to improve system readability.

  5. The data is scaled down to 28x28 pixels to match the training set data.

  6. The values 0-255 are scaled down to 0-1 to match the training set data.

The standardized data processing improved the system’s accuracy from ~50% to >95%.

MEMS Adjustable Bandpass Filter

Device utilizes a system of fixed combs to simulate capacitors and folded springs to simulate resistors.

When a DC voltage signal is applied to the system it moves the combs and alters the bandpass range.

An AC signal applied to the top comb produces a filtered signal and the bottom comb.

Microscope image of MEMS device

SEM image of MEMS device