Daryl Lim
I build AI that runs on your own machine, and agents that show their work.
Independent AI engineer in Honolulu.
What I build
On-device AI. I build private apps that run on Apple Silicon with MLX and mlx-vlm. My largest, granit, turns meeting recordings and documents into transcripts, Markdown and cited answers without anything leaving the Mac. Because they run locally, they keep working through an internet outage and cost nothing per use. Why I build this way → · On-device projects →
Agents that show their work. I build agent systems on Deep Agents and Claude that you can check: a grant writer that quotes evidence for every judgment and flags unknown facts instead of inventing them, and a financial planner whose every figure comes from a tested tool, not the model. Agent projects →
Healthcare AI. I build clinical reference tools around review and privacy: transcription that flags low-confidence words, locks export until a clinician signs off and keeps an audit trail that never records patient data, plus Corti SDK apps for ambient scribing, dictation and medical coding. Healthcare projects →
Featured projects
- On-device AI
- On-device AI
- Agents
- Agents
- Healthcare AI
- Retrieval
Latest posts
- The model never does the math
In my agent projects, every number comes from code, not the model. Here is how that works in a financial planner and a grant writer, and why it reaches further than arithmetic.
- Why I build AI that runs on your own machine
Much of what I build runs entirely on a Mac. Here is why, what it costs, and when I use hosted models anyway.