SignalForge Pay — AI Revenue Decision Agent
Inspiration
Small businesses see payment activity in dashboards, but the difficult question is what to investigate next. SignalForge Pay is designed to turn PayPal transaction signals into an evidence-backed revenue recovery brief that a founder can act on.
What it does
The web dashboard presents gross and net revenue, refunds and reversals, fees, simulated failed payments in demo mode, chargeback event signals, average order value, historical revenue bar charts, anomalies, and a simple next-period forecast. An AI reasoning endpoint turns aggregated evidence into a concise explanation of what changed, hypotheses to investigate, risks and opportunities, and three prioritized next-best actions. It distinguishes observed facts from interpretations and explicitly warns against confusing correlation with causation.
How we built it
Frontend: React, TypeScript, responsive dashboard, Floot deployment. PayPal integration: a server-side PayPal Sandbox OAuth 2.0 flow and Transaction Search API integration using environment-held Client ID and Client Secret, designed to retrieve and aggregate test transactions. No real payments are initiated or processed by the app. AI: Floot AI service using GPT-6 Luna; the model receives calculated metrics and an anomaly summary rather than unfounded prompts. Validation and transport: Zod input schemas and SuperJSON API transport. Demo mode: clearly labeled synthetic financial data. Judges can also download a safe sample PayPal Transaction Search-format JSON file and import it into the browser to verify transaction classification and changes to the dashboard. Imported rows are processed locally; only aggregate metrics are sent to AI after explicit user action, and imported file authenticity is not verified.
What is currently working
The hosted dashboard, local PayPal-format JSON importer, historical revenue visualization and AI Revenue Recovery Brief are functioning. On November 6, the working demo flow produced a real AI response to explicitly synthetic data. Importing a separate synthetic PayPal-format report recalculated the dashboard and cleared stale AI output. The demonstration uses actual application captures with edited timing, English narration and captions; it does not claim to be an uninterrupted live recording.
The November 6 Sandbox verification confirmed successful OAuth authentication but no required reporting/search/read scope in the issued token. The protected sync reports PAYPAL_REPORTING_SCOPE_MISSING, and successful end-to-end Transaction Search retrieval remains unverified. The interface preserves explicit data-source labels and errors; importing a JSON report is not presented as a successful API sync. No real payments or customer financial data are used.
The published source was synchronized with GitHub on November 6, with all 197 Floot source files checked for parity, four Jasmine spec files passing, and a clean TypeScript check. The public app returns HTTP 200; an invalid sync passcode returns HTTP 401. The repository includes the complete Floot-managed application source, README, MIT license and environment-variable names without credentials. Local instructions describe the Floot runtime dependency honestly rather than claiming an independent npm server.
Challenges and lessons
Financial AI is only useful when its inputs are trustworthy. We chose explicit source labels, calculation-first analysis, reversible recommendations, and clear warnings about uncertain causal claims. Our October 21 improvements include conservative, test-covered PayPal event classification (sale, refund, reversal, chargeback, hold and payout categories), mixed-currency safeguards, merchant-only Sandbox sync access and a durable PostgreSQL hourly cap on public AI calls. The reporting-permission limitation is documented. Forecasts are simple decision-support estimates, not validated financial predictions.
Built during the hackathon
SignalForge Pay is a distinct PayPal-oriented app created from October 10, 2026 onward, informed by the previously existing broader SignalForge AI concept. The PayPal transaction architecture, revenue analysis dashboard, demo data model, revenue-brief API and PayPal-specific user experience were built in the hackathon submission period.
What's next
Resolve the Sandbox Transaction Search permission for the existing app and verify end-to-end reporting when that grant becomes available. The current prototype and its working AI/import flow are preserved. The public 2:55 demonstration is available on YouTube: https://www.youtube.com/watch?v=QB8WXaeiY5I .
Working preview: https://signalforge-pay.floot.app
No real payments or customer financial data are used.
Built With
- floot
- oauth2
- paypal
- postgresql
- react
- superjson
- typescript
- zod
