INDEPAIFIRE planning with a city and date
A financial-independence product for digital nomads and ambitious professionals. IndepAI turns "retire someday" into a concrete FI number, target city, and timeline, supported by calculators, AI coaching, geo-arbitrage data, and Supabase-backed account flows.
01 · Why it matters
Most FIRE tools stop at a spreadsheet number. IndepAI connects that number to where a user can live, how their monthly savings gap changes, and what actions move the date forward.
The product is opinionated: show one obvious next action, keep the financial math deterministic, and use AI to explain trade-offs instead of inventing calculations.
02 · What we built
A public acquisition layer with calculators and SEO pages, authenticated dashboards for FI Score and city comparisons, AI coaching flows, email lifecycle infrastructure, analytics instrumentation, and Supabase-backed data models.
The system includes internal launch operations: database audits, Search Console setup guides, email warmup, PostHog funnel validation, and scripted release checks.
03 · Engineering decisions
Astro keeps the marketing and tool pages fast. Supabase holds user, portfolio, and city data. AI calls are wrapped behind explicit prompts, eval thinking, and privacy boundaries.
The strongest product decision is restraint: the app avoids generic "wealth dashboard" noise and keeps nudging users toward a clearer city, date, and savings lever.
How the system is actually put together.
Astro product shell, Supabase backbone, deterministic finance core
IndepAI is built as an Astro SSR application with React islands where the product needs interactivity: dashboards, calculators, portfolio workflows, AI coach screens, city comparison, and planning tools.
The financial core stays deterministic. FIRE projection and portfolio health logic are pure TypeScript modules reused by calculators, previews, dashboards, and AI context generation, while Supabase Postgres stores users, portfolios, snapshots, trips, feature flags, and audit data behind RLS.
- Astro 6 SSR with React 19 islands for dense product workflows.
- Supabase Auth, Postgres, RLS, portfolio tables, nomad planner tables, and pgvector knowledge chunks.
- Pure FIRE and portfolio engines used before AI explanation is allowed into the flow.
Consent-gated ToolLoopAgent instead of free-form financial advice
The AI coach is not a loose prompt box. It runs through a guarded API endpoint with consent checks, PII sanitization, quota and cost ceilings, usage logging, streaming, metrics, and explicit tool definitions.
Tools are bounded to product workflows: city costs, tax and visa checks, portfolio reads, FIRE math, RAG search, approved financial writes, city updates, asset creation, and consent toggles. The agent explains trade-offs using the deterministic context rather than replacing the calculation layer.
- OpenRouter-powered ToolLoopAgent with tier-specific step and cost limits.
- pgvector and full-text knowledge search through explicit RAG tools.
- PostHog, Prometheus, Sentry, and cost ledger events around AI usage.
“IndepAI is the product we use to prove how we think: deterministic core, useful AI at the edges, and launch operations that can survive real users.”
Start with a two-week systems audit.
We map architecture risk, delivery slices, deployment gates, and where AI or automation can help without losing production control.