一只有品的刺猬
scroll to explore
Have the courage to follow your heart and intuition.Steve Jobs
01 / AGENTS
orchestration layer
Building things that think in loops
02 / LLM
post-training
Shaping the model after pretraining
These are the tech stacks that I'm familiar with
03 / FULL STACK
ship surface
From idea to shippable in a weekend
04 / PRODUCTION
ship & keep shipping
Taking AI from demo to reliable
Autonomous incident response: Datadog alert → LLM investigation → auto-fix → GitHub PR.
Production-ready scaffold for full-stack AI agents with CopilotKit + LangGraph.
AI agent that plays Slay the Spire 2 autonomously via pure LLM reasoning.
From one sentence to a coordinated agent team. No dashboards. No servers. Just planning.
You describe your app. It creates the video. AI-powered product video pipeline.
Temporal knowledge graph for personal journals with autonomous optimization.
AI-powered storybook generator with real-time streaming and human-in-the-loop.
The source of truth for LLM-assisted RAG setup. Rich content, minimal metadata.
Weather-based outfit recommendation app with soft minimalist editorial design.
Remote monitoring and control of Claude Code sessions through Telegram.
Turn your daily work into social media posts. X, LinkedIn, Reddit.
Building either dev tool or an actual product.
Currently working with K-12 education AI.