DS/ML Engineer focused on computer vision, neural network training, compression, autonomous driving, LLM fine-tuning, and GitHub projects.
AI Fluency
Article·48m ago
GPT-OSS-20B tool-use fine-tune
I engineered a GPT-OSS-20B tool-use fine-tune using a standard Unsloth notebook and the NousResearch dataset. The resulting model reached more than 23,000 downloads on Hugging Face.
I was flown from Miami to San Diego by the comma.ai team for Comma Hack 7 after submitting a solution to their lossy video compression challenge. I trained two custom neural networks: one identified road boundaries and preserved the road pixels needed for autonomous navigation, while the other calculated the appropriate gray pixels to preserve warped motion between frames. I compressed a 37.5 MB dashcam video to 191 KB and reached the top of the leaderboard at the time, improving the score from 0.187 to 0.172 and doubling the progress of the previous 30 submissions combined.