Projects

02

Prediction markets were taking off and people were claiming you could set up a trading bot overnight. That did not add up to me. I wanted to find out what it actually takes, and whether there is a real edge behind the noise.

I started from first principles: how weather data is measured, where stations source it, and how it gets hosted. I put the system in an AWS data center co-located with NOAA's weather data to keep latency down. It is Python and NumPy. It ingests real-time NWP ensemble forecasts, builds probability distributions with station-level bias correction, and trades on prediction exchanges with risk controls: exposure ramps, drawdown circuit breakers, and per-market caps. It runs on EC2, processing over 200 markets per cycle across 20 US cities, and sends me performance reports.

200+

Markets per cycle

20

US cities

1. 01

   NOAA data

   Real-time NWP ensembles

2. 02

   Bias correction

   Station-level adjustments

3. 03

   Probability model

   Empirical distributions

4. 04

   Trade execution

   Caps, ramps, breakers

03

Creative Pipeline

AI-generated content was taking off, and most of the conversation around it felt like hype. I wanted to understand it from the ground up: not just the end product, but how you actually get there. I took the process apart, figured out how it was made, and rebuilt it myself.

You give it a one-line premise. The system writes a scene-by-scene script, produces voiceover per scene, generates AI-animated clips, and assembles a finished video with word-synced captions, ready to publish.