This is the shortest path through my systems work. The three threads share a common concern: making state, ownership, failure, and resource boundaries explicit enough to reason about.
Distributed Systems
How to Read TiDB Source Code
A five-part path through TiDB’s server lifecycle, SQL execution, transaction handling, and supporting subsystems: Part 1, Part 2, Part 3, Part 4, and Part 5.
The series was written against the 2020 source tree. Its reading method remains useful, but individual code paths and package names may have changed.
How TiDB Implements the INSERT Statement
A code-level walk through planning and executing one of the most common SQL write paths.
OceanBase Internals: Transactions, Replay, SQL Engine, and Unit Placement
A map of how transaction processing, recovery, query execution, and resource placement meet inside a distributed database.
Agent Systems
agentd: A Transport-Neutral Runtime for Personal Agents
An architecture note on turn contracts, explicit capabilities, durable state,
scoped execution, and inspectable outcomes. agentd is experimental and
private; the article describes the design without presenting it as a public
platform.
Running OpenAI Symphony as a Solo Development System
What changes when an agent workflow is treated as an operating system for work rather than a collection of prompts.
Tailgate: A Private AI Gateway
A smaller project note about centralizing provider access, policy, and observability at an AI gateway boundary.
Inference Systems
This is the newest part of the portfolio. The current public notes document hands-on serving experiments; scheduler, cache, routing, and performance work will be added only when it has reproducible code or measurements.
Qwen Local on Apple Silicon
A practical local-serving setup and the operational constraints that surfaced while running a model on consumer hardware.
Exploring Local LLMs with Ollama
An earlier field note on local model serving and application integration.
Projects and Code
I keep exploratory tools and personal projects elsewhere on the blog. This page is intentionally narrower: it collects the work most relevant to distributed systems and AI infrastructure.