Summary
All About AI walks through the complete construction of an AI-powered prediction market trading bot targeting New York City daily temperature markets on Kalshi, combining three recently released models in a single pipeline. The build uses OpenAI’s GPT-6 Astra as the reasoning and orchestration layer, Google DeepMind’s WeatherNext 3 (accessed via Google Cloud BigQuery and Earth Engine) as the meteorological forecast source, and Codex running in “yolo” mode as the coding agent — deployed on an AWS VPS for continuous 24/7 operation.
The tutorial covers every integration step: requesting API access to WeatherNext 3 through Google Cloud, structuring bucket-prediction prompts for Kalshi’s temperature event format, running backtests across opening time windows from +1 to +30 minutes, and authenticating programmatic trading via the Kalshi private key API. Backtesting results show approximately a 13% improvement in prediction accuracy at the opening +1 minute window (64 vs. 74 scoring, where lower is better), though the creator acknowledges the sample is too small to confirm a durable edge.
Beyond the Kalshi-specific use case, the video functions as a practical template for combining large language model orchestration with domain-specific forecasting APIs and financial market interfaces. It also offers early firsthand observations on GPT-6 Astra’s performance in a real coding and reasoning context, including high-reasoning mode configuration within Codex.
📺 Source: All About AI · Published September 08, 2026
🏷️ Format: Hands On Build
