Silicon Valley · Long-form Conversations

In Silicon Valley, we sit down withpeople doing the hard things.

Subscribe on YouTube

Uncle Moon is a technology creator and a believer in long horizons. We document the real paths of researchers, founders, and engineers—less hype, better questions.

THE ARCHIVE

Long-form conversations

Every choice has a context
EP. 0062026.09.08117 MIN

Zhibin XiaoThe Full AI Chip Stack: Manufacturing, Inference, and Startups

Zhibin Xiao traces a career from Zhejiang University and UC Davis to Oracle, Alibaba, and founding ZFlow.AI. We map the global chip supply chain, AI inference bottlenecks, NVIDIA's moat, and the architectural bets behind Groq, Cerebras, and SambaNova.

  • AI Chips
  • NVIDIA
  • AI Inference
  • ZFlow.AI
EP. 0052026.08.23118 MIN

Tianyin XuBuilding Fault-Tolerant Systems for the AI Agent Era

UIUC professor Tianyin Xu discusses award-winning systems research, mentorship, and how AI agents are reshaping interfaces, SRE, and reliability. We also go deep on formal verification, model checking, specification-driven development, and recovery-oriented computing.

  • Agent Infrastructure
  • Systems
  • Formal Verification
  • Reliability
EP. 0042026.08.0865 MIN

Todd LiFrom Google to YC: Building an $80M Voice AI Business

Retell AI co-founder Todd Li traces the path from Google engineer and two YC applications to building an AI voice customer-service company with $80 million in revenue in two years. We discuss repeated pivots, enterprise delivery, five co-founders, competing with OpenAI, disciplined hiring, and execution.

  • Retell AI
  • YC
  • Voice AI
  • Execution
EP. 0032026.07.2785 MIN

Meng ZiliTsinghua at 15, Professor at 23—Then Starting a Company

From Tsinghua and MIT to HKUST, and from network systems and cloud gaming research to founding WiCi. We talk about timing, connection, long-term thinking, and the abilities that matter most in the AI era.

  • Tsinghua
  • HKUST
  • Cloud Gaming
  • WiCi
EP. 0022026.06.08145 MIN

Jiang JunchenKV Cache Is the Next Generation of Memory for Large Models

A long conversation spanning networked multimedia, open-source infrastructure, and entrepreneurship—the ideas behind LMCache and TensorMesh, and a research path from Tsinghua's Yao Class and CMU to the University of Chicago.

  • KV Cache
  • LMCache
  • CMU
  • Open Source
EP. 0012026.05.17106 MIN

Banghua ZhuFrom SGLang to a Second Startup: Knowing When to Walk Away

From Tsinghua, Berkeley, LMSYS, and Chatbot Arena to reinforcement learning, SGLang, an NVIDIA acquisition, and a second startup. We also talk about the courage to choose again at pivotal moments.

  • SGLang
  • LMSYS
  • Reinforcement Learning
  • Startups

WHY WE TALK

More important than the answer is how someone arrived there.

Technology moves fast. Human choices are still worth understanding slowly. We keep the camera on people who create, decide, fail, and begin again at the frontier.

Explore the interview library