I Had Never Built a Mapping App. Then I Built One with Codex and Claude.

I had never built a mapping app. I wanted to answer a deceptively simple question about a road intersection: what is here, and how do we know?

That question became Intersection Intelligence — my personal mapping experiment and internal, non-commercial research tool. Codex helped with implementation and verification; Claude contributed architectural review. I owned the product vision, scope and decisions about what was good enough to ship.

The interesting part was the iteration:

• Learning MapLibre and PMTiles, then building a map journey from South Africa down to a single corridor.
• Turning inconsistent source datasets into usable evidence through custom importers, fuzzy matching and human review.
• Working through repeated rendering and query performance tuning cycles.
• Preserving reviewed human contributions as the underlying map data changed.
• Designing the software and deployment around a small Linux B2 host, then deciding what belonged in an achievable research MVP.

I've shared the journey in two companion posts:

1. The full build story

How I built Intersection Intelligence, with the help of AI agents

The narrative, lessons, redacted South African map screenshots, architecture diagrams, technology stack and deployment decisions.

2. The slides and the story

Building Intersection Intelligence: the interactive presentation

A 28-slide animated HTML talk with navigation, speaker notes and an expandable written explanation of each slide.

My biggest lesson: AI made an unfamiliar stack approachable, but I still had to define what the data meant, test the user experience and own acceptance.

If you've learned a new stack with AI, what took the most iteration — data, architecture, performance, or deciding when to ship?