Deterministic Random Walk

Deterministic Random Walk

Markets, institutions, and AI systems all operate under uncertainty. The practical work is to navigate that randomness without mistaking it for chaos: identify what keeps holding, encode it carefully, and leave room for judgment where judgment is actually needed.

This is the frame behind the name. Investment is partly art because markets are noisy, reflexive, and human. It is partly science because decisions still need process, evidence, controls, and memory. Production AI has the same shape: probabilistic models help where meaning is ambiguous; deterministic engineering makes the result repeatable, auditable, and safe to scale.

This site is a research log on modernizing investment workflows with production-grade AI: how to combine semantic judgment with reliable systems, and how to move through randomness without outsourcing discipline to it.

A fixed rabbit template surrounded by energetic hand-drawn circles.
A fixed template inside wild circles: one small picture of the whole idea.

Featured series

English

  1. Part 0: Why This Series Exists
  2. Part 1: The Impossible Disclosure Task
  3. Part 2: Why One Classifier Became Seven Layers
  4. Part 3A: Where LLMs Shine
  5. Part 3B: The 90% Agents Would Destroy

中文

  1. 第0篇：这个系列为什么存在
  2. 第1篇：投资组合经理的不可能任务
  3. 第2篇：从“直接扔给 AI”到 7400 行代码
  4. 第3A篇：LLM 闪耀的地方
  5. 第3B篇：智能体会毁掉的 90%