Show HN: ride-recap, teaching a LLM my taste to automate cycling highlights (www.iandmacomber.com)

A developer has created an innovative pipeline called ride-recap that automates the process of generating highlight reels from cycling footage captured by a GoPro and enriched with data from a Garmin fitness tracker. Over three months, the system evolved through extensive iteration and uses the Gemini LLM to analyze hours of video alongside telemetry data, identifying and ranking the most visually compelling moments. This process transforms a lengthy ride into a concise 60-second video, showcasing key segments with an overlay of telemetry data, all at a minimal cost of about $0.04 per ride. This project is significant for the AI/ML community as it demonstrates how large language models can be trained to develop a nuanced understanding of "taste" in video editing through iterative feedback and manual labeling of clips. The creator highlights the challenge of teaching an LLM to discern interesting content, revealing that success hinges on continuous refinement of prompts and candidate selection processes. The ride-recap tool not only enhances the personal experience of cycling enthusiasts but also exemplifies the practical applications of AI in creative domains, sparking potential future business opportunities in automated content generation.