IndepAI architecture: deterministic FIRE math with useful AI at the edge

A FIRE product cannot let AI invent the number. Users need a financial independence target, a city comparison, a savings gap, and a date they can reason about.

That is the key boundary in IndepAI. The core calculations stay deterministic. AI helps explain trade-offs, summarize scenarios, turn city and portfolio data into plain language, and guide the user toward the next action.

The architecture reflects that boundary. Astro serves fast public tools and SEO pages. Supabase stores user and portfolio data. Analytics validates which flows work. AI services sit behind explicit prompts and product constraints.

That boundary is the product strategy. FIRE planning has to feel personal, but the system cannot become vague when money, timelines, and city choices are involved. A user should know whether the number came from a calculator, a stored portfolio snapshot, a city-cost assumption, or an AI explanation.

CloudFloo keeps the acquisition layer fast with Astro and focused public tools. React islands appear where the interface needs richer interaction: calculators, dashboards, portfolio flows, city comparisons, and AI coach screens. That lets the product stay fast for SEO without giving up interactive product depth.

Supabase gives the product a practical account and data backbone: auth, Postgres, RLS, portfolio tables, planner data, and knowledge chunks. The important part is not the vendor choice. The important part is that source-of-truth data and AI context are separated enough for privacy, audits, and future product growth.

The AI layer is useful because it sits at the edge. It can explain a trade-off, summarize a plan, retrieve relevant context, or help a user decide the next action. It should not replace the projection engine, bypass consent, or write sensitive data without a bounded workflow.

Launch operations are part of the architecture too. PostHog events, Search Console setup, email warmup, Sentry, release checks, and funnel validation decide whether the product can learn from real users after the first public push.

The result is not "AI finance magic." It is a practical FIRE product with an AI layer that makes the plan easier to understand without taking over the math.