Edward Grigoryan

A university project with the goal to parse photos of Sudoku into their digital form.
I used OpenCV and open-source algorithms to crop the Sudoku grid from the background. Next, using breadth-first search (flood fill), I separated each digit from the grid into separate images. My optical character reader labeled each digit image. I used the handwritten MNIST dataset and a known convolution network to train the model to develop the OCR. After that, I fine-tuned it with programmatically generated (Pillow) printed digits with a dozen fonts. Resulting OCR had a 99% accuracy which translated into 90% accuracy for the entire Sudoku grid.

A research project to detect a drone or plane's location by tracking changes in surface images beneath the drone. This was a backup in case GPS navigation failed for any reason. Using linear algebra and OpenCV algorithms, my position detection algorithm got good results. While it could not detect the drone's precise position, it was enough to direct it to the desired place. The performance of the analysis was also acceptable enough to run on an onboard Raspberry Pi.

I started learning to program when I was 12. After rigorous and persistent training, I became exceptionally skilled at algorithms and data structures and won the silver medal at IOI 2014, Taiwan.

From the IOI website:
"The IOI is one of five international science olympiads. The primary goal of the IOI is to stimulate interest in informatics (computing science) and information technology. Another important goal is to bring together exceptionally talented pupils from various countries and to have them share scientific and cultural experiences."

The client, a signal processing company, needed a LabVIEW API allowing them to listen and transmit on a wide range of radio frequencies using HackRF devices. I developed a cross-language interoperability layer between LabVIEW and HackRF C++ drivers, allowing LabVIEW developers to build seamless and low-latency communication with HackRF devices.

Developed a full-stack demo application for a 2-founder AI startup to pitch their idea.

I designed the website using Material UI and developed the front end that enables users to search for data of their interest with various filtering capabilities. I also developed an Elasticsearch-backed Python back end to handle incoming front-end requests.

The client wanted users to be able to send SQL requests to the database from the front end. The back end should have returned only the data allowed to be viewed by the user, omitting the rest.

WORK DONE
• Developed endpoints to get users SQL requests from the front end to the back end.
• Developed a custom SQL parser to eliminate SQL injection attacks and filter user-related data
• Integrated an SQL parser with SQLAlchemy to translate user-generated SQL queries to safe ORM commands.
• Integrated the most used SQL features like select, join, where, group by, order by, and having 140 SQL functions.
• Modified the front end to work with new endpoints.
• Improved the front-end layout and query-related error reporting.