Skip to content

README.md

Latest commit

 

History

History
65 lines (43 loc) · 3.36 KB

README.md

File metadata and controls

65 lines (43 loc) · 3.36 KB
Raw

FinPulse AI

FinPulse AI is an educational financial-literacy and scam-awareness prototype built for the Work in Fintech AI Summit & AI Hackathon 2026.

It demonstrates:

  • plain-language checks for common high-risk promotion patterns;
  • a guided pre-decision rehearsal with missing-context questions, a personalised downside scenario, a comprehension check and a private Decision Receipt;
  • a simplified UK payslip learning calculator;
  • hypothetical investment stress-test visualisations;
  • gamified financial-literacy lessons;
  • a concept dashboard for potential partner outcome measurement.

Important limitations

FinPulse AI does not provide financial advice, determine whether a firm is authorised, perform real image/video OCR, certify financial capability, or establish FCA Consumer Duty compliance. Outputs are educational prompts produced by deterministic rules. Verify firms directly through the official FCA Register and Warning List, and verify tax results against current HMRC guidance.

90-second judge demo

  1. Select 50x Forex Scam and run Check Claim Risk Indicators.
  2. Keep Maya's example amount and rental-deposit reason, then select Rehearse Maya's Decision.
  3. Show the missing context and personalised leverage downside.
  4. Answer It magnifies losses as well as gains.
  5. Generate the private Decision Receipt and close on the independent-verification next step.

The receipt records educational completion only. It intentionally stores neither the pasted claim nor Maya's personal reason.

The public deployment stores only short-lived, non-identifying demo events in memory. It intentionally has no durable customer database. Partner telemetry routes are hidden unless ADMIN_API_TOKEN is configured. The app does not mirror the FCA Warning List because its former RSS URL currently redirects to a missing page; users are sent to the official live list instead. A production service would additionally require authenticated users, tenant isolation, consent and retention controls, a durable managed datastore, monitoring, regulatory review, and independent security testing.

Local development

Requirements: Node.js 24.

npm ci
npm test
npm start

Open http://localhost:3000. The API health endpoint is /api/v1/health.

To call protected development endpoints, set a long random ADMIN_API_TOKEN and send it as a bearer token. Do not expose this token to browser code.

Deployment

The repository includes a Vercel configuration that routes the website and API through the Express serverless entry point. Vercel creates preview deployments for pull requests and a production deployment from the configured production branch.

Before production deployment:

  1. connect the GitHub repository to the Vercel project;
  2. configure a long random ADMIN_API_TOKEN in Vercel project settings;
  3. require the CI / test check and pull-request reviews on master;
  4. validate the preview URL, including /api/v1/health, before merging.

Security and privacy

  • Only an explicit allowlist of website assets is public; source and data files are not served.
  • API payloads are size-limited and rate-limited.
  • Partner telemetry and exports require server-side authentication.
  • No user-provided claim, salary, or referral data is written to disk.
  • Dependency updates are monitored with Dependabot.

Licence

MIT. Third-party trademarks belong to their respective owners.