Intelligence,
grounded in reality.
Understanding,
built to last.

I’m Lucius He, an independent builder and researcher. I explore how AI can understand physical objects, how groups of agents reach decisions, and how people carry understanding forward.

Across these projects, I build systems, test their assumptions, and keep the evidence available for the next person to question.

Inspect the work

Questions made concrete.

Three directions of work, with software, experiments, and a reading collection you can explore.

Moonblade

Engineering prototype

What would it take for the next person—or the next AI—to inherit an understanding of a particular physical object, then check it again when reality changes?

Moonblade explores how a physical object could carry identity, knowledge, task skills, and context for compatible AI. Its current software prototype records what changes, preserves conflicting evidence, and checks when a task can continue.

Read the case

Eight inputs. A recovery record.

Inspect the executed trace: a reversed orientation triggers recovery; conflicting inputs remain visible; later evidence resolves the conflict.

Input 3
orientation = reversed
Runtime enters RECOVERING.
Input 4
orientation = correct
Recorded as CONFLICTED.
Input 7
orientation = correct
Resolved to KNOWN; advances to s3.

Synthetic software trace. No hardware operation or field success is claimed.

Inspect the executed trace

SwarmAlpha

Multi-agent decision research

How do agents with different, partial views arrive at a collective decision—and what can an outside observer measure?

I built a multi-agent experiment system to study how information disclosure changes reported judgments, alongside tools for observing group processes. The archive preserves working code, model runs, and the findings that changed the research direction.

Read the case

Information disclosure can change a group’s reported judgment.

Experiments compared what agents reported after receiving different information. The results prompted further work on how to describe and measure the group’s changing state.

Explore the research

The Ancestors

Personal cognitive infrastructure · Public reading slice

What can earlier thinkers help us see in the questions we face today?

The Ancestors is my personal atlas of human questions, earlier thinkers’ works, and their competing answers. Alongside its reading paths, I keep a record of how my own questions and judgments change.

Read the case

Am I building the life I want—or the life others have taught me to admire?

A life can look successful from the outside and still leave its owner unsure. Read with Adam Smith, Harry Frankfurt, and John Stuart Mill: what deserves admiration, which desires to stand behind, and what choosing for oneself requires.

Follow the question

Make the next judgment possible.

These projects did not begin as a unified system, and they do not share a technical core. Their questions gradually converged: what must remain visible when intelligence acts, when people disagree, or when understanding passes to someone else?

My approach is to keep the input, the decision, and the evidence close together. A software trace should explain what happened. An experiment should say what was varied and what was measured. A reading path should let another reader return to the source and disagree.

I use AI extensively in research and implementation. I am responsible for the questions, system and experiment boundaries, critical tradeoffs, and acceptance of the work. AI assists with analysis, drafting, implementation, and checks. These pages distinguish an implemented mechanism from an executed test, and both from a result in the physical world.

About

I am Lucius He (贺孟元), an independent builder and researcher completing my high school education.

My recent work brings together software prototypes, multi-agent experiments, and a long-term reading collection. This site presents the projects I most want to develop through testing and conversation.

Earlier work includes Personal Agent Runtime, which separates a model’s proposed action from permission and verification, and the Phase Reversal Hypothesis. RIZHICHI remains accessible as a secondary page.

Let’s compare notes.

I welcome concrete conversations about object-grounded systems, experimental design, and the inheritance of understanding. A counterexample, a reproducible check, or a better question is a good place to start.