I'm Zhengyao Jiang, Cofounder & CEO of Weco AI, and a computer scientist by training.
I believe the scientific method is, at its core, an algorithm: hypothesize → experiment → analyse. As more of our frontier knowledge is encoded in software, from high-frequency trading systems to generative AI training pipelines, the next leap in AI will come from machines that can run the scientific innovation algorithm at scale and encode the knowledge into code.
Weco AI is my bet on that future. Our AIDE algorithm converts the slow, manual “hill-climb” of research engineering into a fast, evaluation-driven search that routinely uncovers non-obvious, high-performing solutions. Think of it as a knowledge-generation engine that learns directly from empirical feedback rather than human hunches.
This focus on AI systems that learn from external feedback traces back to my PhD work at University College London, supervised by Tim Rocktäschel and Edward Grefenstette, where I studied how reinforcement-learning agents can extract maximum value from every interaction with the environment. The same principle now guides everything we build at Weco: squeeze the most insight out of each compute cycle, and let the machine iterate toward discovery.