Episode 15

In this talk, we’ll explore what it takes to run your own LLM locally. Along the way, we’ll unpack the core principles that shape model selection, from attention and the KV cache to post-processing methods and quantization. We’ll examine how these factors help you choose the right model for your use case and hardware, discuss the limitations you’re likely to encounter, and learn how to make informed trade-offs.

Self-driving cars are arriving on public roads, and the same core technology is finding its way into agriculture. At Evopilot, we retrofit tracked carrier machines with a sensor kit and software stack that lets them drive autonomously through apple orchards and vineyards while mulching and spraying. In this talk, we trace how the system evolved from early prototypes to rovers working in the field: what we built first, what broke, and how each round of field testing reshaped the hardware and software. Along the way, we share the engineering challenges we faced, from keeping a tracked vehicle on narrow paths to making it safe to operate around people, and the lessons that shaped today’s design.