I code.

AI products, multimodal systems, and infra tuned until behavior becomes legible.

01 / Work

AI products with infrastructure underneath.

From vague intent to shipped control surfaces.

HA Distributed services

Data Structured / unstructured

Infra Models / containers / tools

Lead/SRE Delivery / reliability

Product Scope / boundaries / delivery

LLM systems LangChain / retrieval / agents / structured output

Generation Stable Diffusion / LoRA / ComfyUI / GPT-SoVITS

Backend FastAPI / Docker / model serving / internal tools

SRE TDSQL / automation / containers / remediation

02 / Lab

Lab / private systems.

Small systems, tools, and experiments. Some private, some forked, all reworked into usable surfaces.

Personal AI OS

Devices as interfaces

devices / voice / events / gateways

Knowledge engines

Memory as infrastructure

schemas / queues / graphs / retrieval

Agent interfaces

Messy platforms, clean tools

CLI / MCP / browser / task I/O

Generative canvases

Pixels under direction

image / voice / timeline / model control

03 / Art Direction

Art is a timeline you can scrub.

Motion, text, shade, and progress on one controlled axis.

frame.current = 000/144 0.00s / 6.00s

pixels.bind(bit)

ScrollVideo study

Scroll becomes time.

Frame, annotation, and exit state share one progress value.

<Timeline source="scroll" /> Map gesture to frame

<Control surface="pixel" /> Paint annotations

<Exit when="complete" /> Release the page flow

0% / 0.00s

04 / Shipped

Proof in products, not pitch decks.

PageOn

AI PPT generation: visual representation language, structured LLM output, multimodal agents.

app.pageon.ai

Cyber Space

App Store multimodal AI chat: text, voice, image input, Stable Diffusion, voice conversion.

App Store

Tencent SRE systems

TDSQL, Python ops packages, single-machine deploy, alert remediation, container optimization.

Automation / container optimization / SRE

Public traces

Repos and images: forks, prototypes, containers, and model-service experiments.

GitHub Docker Hub