Index pointer / 001
Journal / latest
A deep dive into Claude Code's open-source codebase, covering the three-level paradigm shift, the agent loop with 4-stage context compression, streaming parallel tool execution, 66+ built-in tools across 6 categories, the dual-call skill system, four-type memory taxonomy, and three multi-agent collaboration modes.
A detailed walkthrough of every major post-training paradigm for large language models, from basic SFT through RLHF, RLAIF, DPO, ORPO/IPO/KTO/SimPO, reasoning RL, Constitutional AI, and tool-use training — with formulas, trade-offs, and the modern multi-stage pipeline that leading models use in 2025-2026.
A deep look at three converging AI trends in 2026 — from half-duplex chat to full-duplex real-time interaction, from multi-modal input to world understanding, and from semantic reasoning to predicting and acting in the physical world.
Aftersun, Alcestis, and philia helped me understand love as attention, reciprocity, and the courage to let another person remain fully themselves.
Index / projects
A long-horizon agent system for enterprise and office workflows, supporting complex tasks, business system integrations, browser operations, scheduled jobs, and multi-skill execution.
An AI&AR empowered chatting application, collaborate with IBM UK.
Read-it-later tool integrated with GPT and Langchain.
A React and Ink coding CLI with branching ReAct, governed tools, Mempedia long-term memory, reusable skills, and sandboxed local execution.
An open-source toolkit for packaging, validating, injecting, and orchestrating reusable multi-agent workers across coding hosts, currently centered on Claude Code.
Pointer / about
This is where I keep the ideas, habits, sounds, experiments, and small changes that are shaping my life—not only the things that fit on a résumé.
Read more about me
