I am Haichang (Charles) Li, a researcher and builder
. Previously, I was a founding member of Sounding.AI
, which is a startup raising 10M+ CNY to explore AI Agent systems. I also worked with Flowtica
and MiraclePlus
(formerly YC China), and collaborated with MEMO
. Currently, I am advised by Prof. Zhicong Lu, and was advised by ACM Distinguished Scientist/IEEE Fellow Prof. Yung-hsiang Lu during my undergraduate studies. My experiences span the real world and academia, shaping my passion for making something people want and a research focus oriented toward practical applications and productization ('From Paper to Product'
) in THE real world.
With a 0-to-1 background spanning AI Startups, Research, PM, and VC, I focus on the gaps between Research, Product, and Market. Previously, my work received the CES AI Best Innovation Award and the Purdue DUIRI Fellowship, with features in Yahoo! Tech and Tech Xplore.
I am glad you are interested in my world. However, this homepage has ceased updates. Until I find the true answer, or create new work that I am genuinely proud of, this space will remain silent.
The following content represents my thoughts on Research—as a Researcher. If you are interested in Charles as a PM/Builder or Charles as a VC, my work in products and at MiraclePlus continues to iterate along established trajectories.
The world moves fast, but sometimes we need a period to slow down.
I am currently navigating a period of foundational refactoring.
My early training was rooted in Human–AI Interaction. Initially, when model agency was limited, I focused on making AI an "Iron Man suit" for humans—using agentic systems to understand user intent for tasks.
However, I have moved beyond surface-level Interaction.
I resonate with The Bitter Lesson — "We want AI agents that can discover like we can, not which contain what we have discovered."
My goal is to grow AI from a supportive "suit" into a true "Iron Man"—an autonomous entity capable of understanding personalized intent and context to take corresponding actions. I want to understand and design such AI to drive the Human-AI relationship from the bottom up, beyond the interface.
Based on this vision, I have begun to re-examine the current research paradigms:
I am questioning whether Human-AI Interaction remains a viable field for scientific research (compared to the raw momentum of industrial data flywheels and real-world users). I question what kind of help users truly need to advance the Human-Machine relationship in this era.
To me, interaction design that remains stuck at the interface level—following standardized lab workflows of design, implementation, analysis, and paper writing—seems devoid of meaning and fails to push the cognitive boundaries of humanity in this era.
My definition of research is pure: it is about solving existing problems through exploratory discovery to expand human cognition. A valid problem must be: 1. Rooted in genuine human needs, and 2. A vital link in a vast chain of evolution. I remain skeptical of discoveries that are not based on real needs or offer no path for subsequent exploration (claiming to be terminal). Therefore, I am rethinking everything.
As part of this refactoring, in the short term, I am exploring:
- How to build collaboration channels for Humans enhancing AI (not just AI enhancing Humans), along with corresponding frameworks for Control, Alignment, and Trustworthiness.
- How AI can leverage Human Preferences and Real-time Interaction Data to achieve richer Reasoning and Agentic Action.
- What kind of Infrastructure we need to build outside the model, rather than relying solely on model training. Models update too fast, but the Paradigm of Interaction can be universal.