Knuth–Plass chunker to fix my obscurely broken semantic search
A change revealed my chunking was quietly wrong. The Knuth-Plass algorithm helped me fix it.
Stories & Processes
On the Blog
I use this space to document how I build, learn, and experiment. These posts unpack the decisions, trade-offs, and processes behind the work.
Benchmarking to show me I'm wrong
How running benchmarks had shown me that I was running on a lot of false assumptions.
LLM Hallucinogenics
What do we now know about why LLMs hallucinate?
Why I turned down an RPA project
Doubts about RPA HIPAA compliance made me turn down a project. Though in hindsight, I changed my mind.
Lessons from my Ambient Clinical Documentation System
What I learned building a clinical AI scribe, and what I would build differently now
For engagement sake: Ship one thing
Shipping one seemingly simple thing that works beats shipping many things that might. The proof is that your users start engaging.
The growing attention deficit
Everyone is out for our attention, more than ever. And we're all disengaging with it all as a result
Building local AI in a sandboxed world
What trying to build a fully local semantic search plugin for Obsidian taught me about constraints, sandboxed environments, and real engineering trade-offs.