Padalo: AI-Powered Remittance Intelligence

AI-powered remittance platform that helps overseas workers send money smarter with provider comparisons, exchange rate forecasts, and shared family budgeting.

Comment

Padalo: Shared Household Finance with Transparent AI Guidance

Inspiration

Millions of overseas Filipino workers make daily sacrifices to support their families back home. The hard part usually isn't sending the money itself, it's staying on the same page afterward: what arrived, what got spent, what's left, and what needs attention next.

As Filipinos, we've lived that fragmented coordination between the person abroad and the family managing things at home. We wanted to build something that treats remittances as a shared household concern rather than a one-way transaction, and that's honest about what it can and can't predict.

That idea became Padalo, a shared household finance workspace for OFW families, paired with a provider-behavior forecasting companion called FXPilot.

What it does

Padalo gives an OFW and their family one calm, shared picture of their finances:

  • A shared household ledger for envelopes, transactions, remittances, and recurring bills.
  • A conversational AI agent that can check budget status, look up upcoming bills, search transactions, log expenses, and create remittances, with a visible tool-progress timeline so the family can see what it's doing.
  • FXPilot, which forecasts provider-behavior patterns (not a tradable FX rate) for Wise, Remitly, and WorldRemit using Prophet, with weekly seasonality, Philippine holidays, and payday/promo effects, always shown alongside a naive baseline and a synthetic-data disclaimer.
  • A deterministic Santos Family demo with an opt-in Demo Mode that reseeds sample data without touching anything else.

Padalo isn't trying to replace remittance providers. It's a neutral layer that turns "did the money arrive, what's left, what's next" from a scattered conversation into a shared workspace.

How we built it

  • Next.js (App Router) + TypeScript for the dashboard, with loading, empty, error, and optimistic states.
  • FastAPI as the API boundary, with SQLAlchemy 2.0 + Alembic for typed, durable ledger services.
  • Neon PostgreSQL for production-like persistent data (SQLite fallback for local demos).
  • OpenAI Responses API driving a typed agent router: the model never touches the database directly, it can only call strict, Pydantic-validated tool schemas, and results stream back over SSE.
  • Prophet for FXPilot's per-provider forecasts, trained on deterministic synthetic AE-PH data, evaluated against a naive same-weekday baseline over a 56-day holdout.
  • TanStack Query, React Hook Form, Zod, Tailwind CSS, Radix UI, Lucide, Playwright rounding out the frontend and testing stack.

Challenges we ran into

Real remittance pricing data is inconsistent across providers, so we built FXPilot on clearly labeled, versioned synthetic provider history instead, with a visible disclaimer on every forecast so it's never mistaken for a live quote or financial advice.

We also had to be careful about trust boundaries around the AI agent. Rather than letting the model reach into the database, we routed everything through typed tools with validated inputs, and surfaced a live tool-progress timeline so the family can see exactly what the agent looked at before it answers.

Accomplishments that we're proud of

We're proud that Padalo went beyond a comparison tool into a full shared household workspace, with a working typed-agent architecture, a real forecasting model with baseline comparisons instead of a black box, and a demo experience deterministic enough to walk a judge through in three minutes.

What we learned

We learned how much of AI product trust comes from architecture, not prompting: keeping the model away from raw database access, validating every tool call, and being explicit about what's synthetic versus real all mattered more than any single prompt tweak. We also learned a lot about time-series forecasting with Prophet, and about designing collaborative tools for people coordinating finances across distance.

What's next for Padalo

  • Real remittance provider integrations and live exchange-rate data.
  • Authentication and production identity verification.
  • Receipt parsing.
  • Migrating FXPilot from synthetic to real provider history.
  • Multi-currency support and mobile apps.

Our goal stays the same: help every overseas worker make every remittance go further, with their family in the loop the whole way.

Built With

Share this project:

Updates