A step-by-step guide to building a production-ready blog from scratch using GCP's $300 free credits. Covers Wagtail CMS setup, Cloud SQL Auth Proxy configuration, TinyMCE editor integration, Docker containerization, Cloud Run serverless deployment, and custom domain configuration with automatic SSL.
April 10, 2026
A practical guide to customizing a Wagtail CMS blog: separating content by audience, sourcing and resizing NASA space images, and overriding the default theme styles.
April 11, 2026
The Complete Hands-On Playbook — From Plan to Production. Covers GCP project setup, Cloud SQL schema design with 3NF normalization, MariaDB-to-Cloud SQL migration, BigQuery dataset creation, and ai_agent_db ERD design for an Autonomous Marketing Agent.
April 23, 2026
Building a Real-Time Data Pipeline on GCP: Datastream CDC, Dataform ETL, and Cloud Workflows Orchestration — From Cloud SQL to ML-Ready Feature Store
April 28, 2026
Predicting top customers with Vertex AI Random Forest, retrieving brand knowledge via LangChain + Pinecone RAG pipeline, and generating personalized Mother's Day offers with Gemini API — From ML Prediction to AI-Powered Campaign Offers
May 2, 2026
Close the loop on a fully autonomous GCP marketing agent — visualizing campaign results with Looker Studio connected directly to BigQuery, and seeing AI Agent in action as Gemini API automatically generates personalized HTML email content for each target customer, delivered via SendGrid on Cloud Run, with full email content and delivery history stored in BigQuery.
May 3, 2026
A first attempt at Business World Models: using TFT + GDELT news signals + Monte Carlo to simulate disaster-driven retail demand shocks across US Polar Vortex 2014 and Germany Elbe Flood 2013. Cross-country validation included.
May 16, 2026
Can AI prevent the next Texas blackout? We build an Energy World Model using real PJM/EIA data, Grid2Op simulation, and a PPO reinforcement learning agent that survives 2x longer than baseline under disaster conditions.
May 22, 2026
Can AI predict rush hour before it happens? We build a Traffic World Model using 6 years of real I-94 sensor data, a Temporal Fusion Transformer that forecasts the full next 24 hours in one pass, and a Gemini LLM agent that converts predictions into actionable traffic advisories
May 24, 2026
Building a physics-consistent world model for robotic apple harvesting, where no simulator and no outcome labels existed. Part 1 covers the problem, the market gap, and the architecture that follows from having to build the environment before training anything.
September 13, 2026
No simulator existed for apple picking, and no outcome labels existed in any public dataset. This post covers how a MuJoCo physics environment was built from scratch, how 31,960 counterfactual outcome labels were generated from it, and why two models were needed instead of one.
Every number in this project rests on labels a simulator produced, so the plans were run back through the physics with the prediction model excluded from scoring. This post covers the four layers of validation, the six silent defects that surfaced, why surrogate fidelity depends on the operating point, what a 500-tree block projection actually says, and what a live installation still needs.
September 14, 2026