An essay on what AI tools get wrong after the first generation. Agreeing on a data structure before working with an agent is a contract — it decides where each change belongs and what stays editable on the second edit, the third, and the handoff. It comes with a gallery of five structural patterns, and the source is on GitHub.

An open-source Agent Skill I vibe-coded with Claude — it builds living customer journey maps from plain conversation, then renders them for any audience: storyline, one-page brief, engineering handoff, or interactive HTML. See the showcase.

A Figjam plugin for designers to collaborate with AI on bulk text editing. Built with Claude in a week.

A Figjam template for designers to qualitatively evaluate and benchmark different LLM prompting methods or agents. Open to all in Figma community.

A two-hour hands-on workshop for UX designers to learn how to design CLI interaction features on input, output, error-handling. It also includes a little bit LLM and AWS resources. Built to help other designers prototyping CLI interface.

A self-serve toolkit for designers to: 1. Analyse recent customer feedbacks. 2. Understand the context when customer submits a comment. 3. Understand cross-console traffic patterns at scale. The toolkit handles data cleaning, data exploration and supports data visualization via locally hosted Streamlit app.

To help the design org's transition to Figma, I'm working on several Figma plguins (swap styles, insert default layers, update to latest design component) to help the team accelerate design process, since external Figma plugins are banned.

2020 version: Wrote a simple internal webpage that parses AWS Transcribe's JSON output, including functions such as easy speaker name replacement. Adapted by other researchers to automatically transcribe audio content.

2024 version: With Gen AI, I'm working on a new version where people can manage multiple conversations and chat with LLM.

On my way relocating to Seattle, I built a Raspberry Pi system to collect GPS info and cabin temperature. This device is connected to AWS IoT core via a mobile hotspot. Then the data is stored in DynamoDB based on the rule and connector I set.

This interface I built allows creators to search components, guidelines, best practices from different public available design systems. For example, table, copy, or links. Further more, user can sign up and save a list of design system pages that he/she is interested in.