Recent advancements in foundation models like CLIP, ChatGPT, and DeepSeek have unlocked a new level of capabilities, reaching a human-level understanding in a wide range of tasks. But transforming these raw models into high-performing, real-world solutions takes more than just API calls.
‍
This talk explores how to use CLIP, ChatGPT and the like for specific use cases, exemplified on the use case of fashion discovery. We’ll start with diving into strategies for curating and generating the right training and evaluation data - whether through interaction signals, human feedback, or LLMs as judges. And we’ll address the MLOps challenges such as deployment, measuring quality in production, as well as iteratively making improvements.
By the end of this talk, you'll have a well-defined roadmap for going from tinkering with foundation models to using them for impactful, real-world advancements.