Sai Coumar

$ Sai Coumar

~ Hi everyone! I'm Sai and I've just graduated from Purdue University with a Bachelors of Science in Computer Science and Data Science. I'm currently applying to become a PhD student in Computer Science with a focus in applied machine learning. I'm primarily interested in Artificial Intelligence for Unmanned Aerial Vehicles (UAVs) and Machine Learning for Image Processing, but I also do some stuff with classical AI and game theory. I'm also a big fan of retro gaming and emulation and I love Final Fantasy!

$ News

Date	Headline
2024-05-17	My medium article "How to Use a Smartphone as a Webcam with OpenCV" has just reached 500 views! Check it out @ https://medium.com/p/b68773db9ddd
2024-05-12	I've uploaded a few datasets I've curated in personal projects to kaggle. Check them out @ https://www.kaggle.com/redsilhouette/datasets!
2024-05-10	I've officially graduated from Purdue University with my Bachelors of Science in Computer Science and Data Science.
2024-01-06	My poker framework for testing AI algorithms is now complete! Check it out @ https://github.com/saiccoumar/Poker.

$ Currently Working On:

~ I'm currently studying transformer model architecture and it's applications in non-language contexts

~ Transformers are so fun and are the backbone of every generative language model right now but I have a sneaking suspicion that the applications it has towards tasks that are less language oriented are severely undercut by the hype around language models. I'm hoping to really dig deep with the Transformer architecture but so far the research has been difficult to interpret...

~ We might be seeing a new iteration of ASCII converter soon!

$ Pinned Project:

~ The one-year anniversary on my ASCII converter is coming up so I want to highlight it for a bit! This is hands down my favorite project of all time.

~ This program would take in image input and convert it to ASCII art which would get outputted directly into the terminal. I was even able to achieve real time conversions for live input or video input as well, with or without color, as well since the algorithms were quite efficient.

~ I'm considering revisiting this using generative deep learning to take a more...overengineered...approach to this task. I'm still yet to see something as fleshed out as I imagine and I think it could come up with some really interesting results.