Flight log
Blog
Write-ups on what I build: how it works, what broke, and what I'd change.
RSS feed · 10 entries
- 01
Craft, curiosity and the common thread in my projects
I am artistic, crafty and curious to a fault. Why that shows up in the software I build, and how trying new things keeps changing what I make.
- 02
Playing sports too hard, and what the injuries taught me
I put a lot of effort into sport and get hurt more often than I should. Notes on effort, recovery and measuring strength, from someone who keeps learning the lesson.
- 03
Hosting a demo of something that cannot be hosted
GraphSentinel needs a Python API, a graph database and a frontend. Here is how I turned a real run of it into a static site anyone can click through, and why I labelled it a replay.
- 04
Tea, in quantity
A short and slightly overdue defence of the drink I get through more of than I should admit.
- 05
About me, in no particular order
Who I am outside the commit history: where I am, what I build, what I do with the rest of my time, and where to find me.
- 06
Welford's algorithm: mean and variance in one pass
How to keep a running mean and variance without storing history, why the textbook shortcut fails, and how I use it to flag unusual option flow.
- 07
Rebuilding this site as a quiet place to read
Why I stripped the portfolio down to one column, and how the blog, case studies and search fit together.
- 08
ChainDeck: turning an option chain into signals
How a raw NSE option chain becomes Max Pain, gamma exposure, volatility panels and a market regime, and why prefix sums, Welford's algorithm and a boring storage layer do the heavy lifting.
- 09
Unbake: recovering spreadsheet formulas from plain values
What happens when a system exports numbers with the formulas stripped out, and how RANSAC, snapping and a blame solver can work backwards to the formulas and the typos.
- 10
GraphSentinel: an investigation agent with a short leash
Designing an LLM agent for security work that can look at a lot and do almost nothing on its own, with typed tools, deterministic scoring, human approval and an audit trail.