盛颖 / Sheng Ying
E247｜对话盛颖：xAI，Infra的浪漫，SGLang，开源，平权与“甄嬛传” presents Sheng Ying as co-founder and CEO of [[RadixARC|Redix ARK]], the researcher behind SGLang, and a former [[XAI|xAI]] engineer who worked on inference infrastructure around Grok. Her story connects Formal Verification, [[SMTSolver|SMT solver]] research, large-model serving, open-source community work, and founder responsibility.
The episode emphasizes motivation and fit. Sheng says she performs badly on work she cannot care about, but can enter intense flow when a problem feels real. That self-knowledge becomes part of her Research Taste: she leaves a mathematically elegant but narrow verification route for AI infrastructure because inference systems feel closer to real deployment pressure.
Sheng’s public position in the source is not only technical. She links Open-Source AI Democratization to her own learning path, argues that infrastructure should be treated as product craft through AI Infrastructure As Product, and frames [[LMSYS|LM-SYS]] as a long-term institution for giving non-established builders and researchers access to high-leverage AI projects.
Connections
- SGLang, Radix Attention, Day-Zero Model Support, and Agent Inference Workload - inference-engine work associated with her.
- [[RadixARC|Redix ARK]] - company she is building around AI infrastructure.
- [[XAI|xAI]] and Grok - early-company work context in the source.
- [[LMSYS|LM-SYS]], [[LMArena|LM Arena]], and [[LianMin|连敏]] - open research community and democratization branch.
- Clarke Barrett, [[StanfordUniversity|Stanford University]], [[ColumbiaUniversity|Columbia University]], Google, and Two Sigma - academic and pre-founder contexts.
- Research Taste, AI Infrastructure As Product, Open-Source AI Democratization, and Gendered Creator Confidence - concepts the source uses through her experience.