RoboRanger
AI/ML Engineering
Won Best Edge AI/ML at a hackathon by building an offline edge AI device that uses CNN-based image classification to identify local species and a voice AI agent to provide interactive, park-ranger-style tours of their ecosystems.
This was my first time doing anything complex on an edge device. I also learned a lot from working with Computer Engineering majors. Biggest lesson learned for working with edge devices: read the specs. And if you’re using LLMs, make sure your LLM has direct information access to the hardware specs as well. We wasted maybe like 2 hours trying to deploy the model on our edge device before realizing it doesn’t even have an FPU (processors built specifically to process floating-point numbers), and our model wasn’t quantized. Somehow, our LLMs never caught that. It was thanks to Hao Luu that we realized our mistake.