How it's built

WavePoint is built by one developer with AI in the loop — 827 commits across the web app and a shared computation layer. This page describes the architecture as it actually is.

The shape of the codebase

The web app is Next.js 16 on the T3 stack — React Server Components, Supabase auth with PKCE, Drizzle, Stripe. Underneath it sits a shared computation layer holding the astrology, numerology, and sacred-geometry frameworks, so every page agrees about where the planets are and what the geometry means.

Computation, not content

Nothing here is served from a lookup table of horoscope text. Every chart, transit, and map line is computed live from astronomical positions — the same engine you can try on the astrocartography demo without an account. Computation modules are pure functions: typed inputs, typed outputs, no session or database access inside. Validation lives at the boundary in Zod schemas. Deterministic, so cacheable by design.

The primitive ladder

The computation layer is organized as an abstraction ladder, modeled on how Stripe layers PaymentIntents under Checkout under Payment Links. Each level composes the one below it.

The interesting work lives at L3 and above — scored and composed layers with no equivalent in existing astrology tools. Each rung is a pure function the pages compose directly, so the computation stays decoupled from any single screen.

Working with AI

Day-to-day engineering is AI-assisted: a developer directing Claude Code inside conventions checked into the repository. Cross-cutting rules load automatically by file path; every directory that needs context carries a short, budgeted brief; and repeatable workflows are committed as skills. Isolation for larger changes comes from git worktrees. The point is a tight human-and-AI loop, not autonomy — the developer stays in the driver's seat.

The heavier framework this repo once grew — initiative governance, a Planner/Worker/Judge task pipeline, behavioral agent roles — outgrew this codebase and became Sherpa, a framework for running Human+AI collaborative workflows. WavePoint now runs the lighter-weight workflow above, and is Sherpa's first customer.

Privacy as architecture

Anonymous computation is the default. The birth chart and astrocartography demos run without an account, and computation inputs are never logged, cached, or retained beyond the request. Birth data is stored only for authenticated users who opt in.

See for yourself

The source is on GitHub, the live product starts at the birth chart, and the person behind it is at robabby.com.