PERSONAL FIELD NOTES

Building, measuring,
and questioning AI systems.

Notes on model serving, inference performance, and the practical edges of modern AI infrastructure.

↓

RECENT NOTES

Learning in public, with evidence.

01 / LEARNING MAP

REINFORCEMENT LEARNING

From Bellman equations to agentic RL: a rigorous learning roadmap.

A visual, paper-linked route through fundamentals, deep RL, world models, offline RL, RLHF, and verifiable-reward agents.

2026
SEPT 28

Next: DQN stability and replay.

Source-controlled, linkable, and evidence-first.

SELECTED WORK

INFERENCE

Reproducible vLLM experiments and system notes for Kaggle T4 GPUs.

RESEARCH

Policy-gated Memory-RSI experiments on public and synthetic data.

REINFORCEMENT LEARNING

从原理课开始，附名词速查、论文路线与可审计实验进度。