I'm Alexiy, a sophomore at Purdue majoring in CS and Math, currently interning at Armada AI, with an unbreakable caffeine addiction (even full-timers at SpaceX were concerned...).
I love learning anything new and learn best by getting my hands dirty. For me, the best feeling in the world is when something works/clicks after bashing my head against a problem for hours or even days.
In my free time, I touch grass (i know, shocking news). I love hiking / mountain biking, and finding excuses to visit coffee shops.
- Favorite Quote: "If you are not in pain, you are not learning." - Employee at SpaceX
- Second Favorite Quote: "Oooooh yea I dont like this problem. It came off a graduate qualifying exam. I put it on there just to torment people, even I don't know how to do it" - MA 453 Prof
- Favorite Coffee Shop: Mount Currie Coffee Co in Whistler B.C. (Philz Coffee is close second)
- Favorite Hike: Sahale Arm / Cascade Pass in North Cascades National Park
- Foreign Languages: Ukrainian, Russian, a bit of Spanish
- Favorite Book: The Martian by Andy Weir
Here's a summary of things I've worked on, feel free to reach out with any questions or just to say hello!
π§ TE AI Cup github repo | dagshub repo | 4/19 Robotics and Intelligent Systems Expo Presentation
- Build standardized, trackable, reproducible framework with MLFlow and DVC
- Use and compare performance of LSTM, time-transformers, DeepESN, temporal fusion transformers, seq2seq at predicting time-series data
- Compare model performance from introducing new data features, changing data augmentation, and adding external indicators
- Final result: Achieve 83% accuracy at an 18-month prediction and 81% accuracy at a 76-week prediction
πPurdue x Google ML full project | maskformer configs | maskformer dataloaders, augmentation, video inference
- Build data pipeline for Maskformer and Mask2former using Google Deeplab and Tensorflow
- Generate, decode, and load TFRecords for panoptic segmentation from COCO dataset with Bash and Python
- Apply random-cropping and color jitter to images and masks, create project config and dataloaders
- Enable running inference on videos
π€ Autonomous Robotics Club of Purdue full project | reworked rocket league sim | wiki
- Improve performance by randomizing training parameters using Rospy libraries and ROS on the autonomy stack
- Refactor simulator to better reflect real-world conditions by randomizing game object physics dynamics, tuning car properties, and simulating latency
- Implement multi-agent training
πΆ BoilermakeX Purdue Hackathon 1st place Dagshub Winner dagshub repo | Wiki |
- LSTM and seq2seq model that predicts specified air quality metrics using input and output features
- Built using Pytorch, with DVC and MLflow to allow for easy testing and augmentation
π Autograders cs240 | cs180
- Used Perl and Bash scripts to run testcases on homework for CS240 and CS180
- Prevented me from getting 0's on hw because of using "\t" instead of "Β Β Β Β Β Β "
πΏπβ Tensorflow Rock Paper Scissors github repo
- Achieve >90% accuracy with 32k parameters for playing rock-paper-scissors
- Use a simplified VGG-19 architecture with data augmentation and inference on live images