Post-Training Is How You Keep Your Taste | Fireworks CEO Lin Qiao

Post-training serves as a critical strategy for businesses to move beyond generic, off-the-shelf APIs and establish a sustainable competitive moat. By encoding unique domain expertise and product-specific judgment into models, companies achieve superior performance while significantly reducing operational costs. This development lifecycle typically progresses from initial prompting and RAG to supervised fine-tuning, preference tuning, and reinforcement learning. Success hinges on high-quality data curation and rigorous, systematic evaluation, requiring deep collaboration between product teams and machine learning engineers. Companies like Cursor and Doximity demonstrate that owning one's intelligence—rather than renting it—enables specialized, cost-effective scaling. Ultimately, post-training transforms AI from a generic utility into a bespoke asset, allowing organizations to maintain their unique product identity and avoid the financial pitfalls of scaling on standard, expensive models.