AI product engineer building tutoring agents and tools for agent workflows.
I work across agent behavior, memory, retrieval, evaluations, backend infrastructure, and product interfaces.
At Pandai, I’m building P&AI, an open-source tutoring agent for students in Malaysia and Brunei. I like working with early-stage teams, testing ideas with users, and improving products through feedback.
- 🤖 P&AI: An open-source tutoring agent for curriculum-based practice
- 🖥️ Ernie: An experiment in agent interfaces that follow the work across sessions, repositories, and Git worktrees
- 🌐 browser.jr: An attempt to rethink the browser engine around agents
- 🌱 Garden: Run agents built for Vercel’s Eve framework on your own infrastructure
- 🦊 Agent Lite: The
agent-browserworkflow, rebuilt around Camofox in an experimental fork - 🧰 Skills: The ways I work with agents, written down to reuse and share
- 🐒 Monkey: Create and move between Git worktrees with one shell command
- 🚢 Ship LLM: Calling an LLM should feel like calling a function (archived)
- 📈 Capstone: Property valuation tools built as executable Marimo notebooks
- 🔎 Retrieval demo: An interface for inspecting retrieval results and their sources
- 🧪 Simkit: Turn descriptions of processes and resources into SimPy simulations
- 🧠 RLM #40: persistent multi-turn environments: Added persistent environments across agent turns, with state history, cleanup, and regression tests
- In defense of Electron: The desktop software Electron made possible and what its successors should preserve
- The best software is yet to be made: Ernie and task-shaped interfaces for recursive agents
- Enduring agents, disappearing machines: Separating the lifetime of an agent from the machines it uses
- I make music and software: Stage fright, listening as craft, and learning to make before feeling worthy