Xuecheng Jiao

While foundation models have demonstrated broad potential in protein engineering, their generic nature often falls short at the industrial "last mile"—where narrow tolerance windows, non-natural substrates, and harsh process conditions prevail. To adapt to these challenges, we integrate decades of enzyme engineering experience with proprietary high-fidelity domain-specific datasets, powered by our CFPS‑driven high‑throughput screening platform (>100× throughput over traditional directed evolution, enabling precise domain-specific annotation). This approach enables us to build practical vertical AI models that have been validated in real‑world production settings, demonstrating improved accuracy over general‑purpose models and paving the way toward an integrated Enzyme Co‑Pilot for reliable, data‑driven biocatalysis.