Melbourne, AU
Solution Engineering · Building · Writing
_
11
Articles published
4
Topics covered
2
Live tools
2
Projects in progress
Sam Bessey: I make modern AI legible for practitioners and the curious. RAG, MCP, fine tuning, quantization in plain English, with working code and real diagrams. Also building Doppio, a live data layer for Google Slides.
Writing & Work
sambessey.com / articlesAI for the Curious — Series
A field guide to modern AI. Working diagrams and real code over buzzwords. My notes on how I explain these concepts to practitioners and the curious.
Perspectives
Top Trumps: LLM Edition
Local models are printing impressive benchmark scores against models ten times their size. Let's read the card and see what happens when you actually play it.
Build vs Buy: The House of Mirrors
Capable open-source AI models have reignited the build vs buy debate. The economics are less obvious than they appear.
Labs
Field report
Fine-Tuning: Training The Actor
Fine-tuning explained with a fun, interactive demo of changing a small model's behaviour to write like an unbearable person on LinkedIn.
Field report
KV Quantisation: Measuring Coastlines
I ran 170 tests to try and break TurboQuant KV compression. Here's what actually failed, what didn't, and why the answer to both questions matters more than the score.
Setup guide
pgvector Setup for RAG
A practical guide to setting up pgvector on Neon and running your first similarity query
Currently Building_
Doppio
A real-time content layer for Google Slides. Doppio connects live data sources directly to presentations — so your deck is always current, no copy-paste required.
{
"slide": "Q2 Revenue",
"source": "BigQuery",
"refresh": "realtime",
"status": "live"
}