Harper Carroll brings practical experience to a growing gap in AI literacy. After working as a machine learning engineer at Meta and briefly at a GPU compute startup in late 2023, she identified a critical problem: almost nobody understood how to fine-tune open source models effectively.

That observation led her to start writing and teaching materials on the topic. Her approach reflects a broader reality in AI adoption: technical skills matter far less than understanding what you're actually trying to accomplish. The tools exist. What's missing is foundational knowledge about how to use them properly.

Carroll's background spans the full stack of modern ML work. Her Stanford computer science degree provided the theoretical foundation, Meta experience exposed her to production-scale systems, and the startup revealed the commercial gap between available technology and practical competence. This combination positions her to bridge the disconnect between what AI engineers build and what organizations actually need to deploy.

The timing matters. As open source models proliferate and fine-tuning becomes more accessible, organizations face a talent problem. They have access to powerful models like Llama, Mistral, and others, but lack people who know how to adapt them for specific use cases. Fine-tuning represents a middle ground between off-the-shelf models and building from scratch, but it requires understanding when to fine-tune, how much data you need, and which techniques work for your constraints.

Carroll's teaching efforts address this directly. Rather than selling compute or models, she's selling competence. That's a different market. Organizations increasingly realize that their competitive advantage comes not from the model itself but from how they apply it to their specific data and problems. A company with mediocre models but strong internal understanding beats a company with state-of-the-art models run by people guessing at parameters.

Her work suggests the next bottleneck in AI adoption isn't hardware or model availability. It's human expertise. As commodity access to powerful models continues,