Check a scam receipt on the phone — NRIC, phone, and email stripped on-device; only scrubbed indicators leave (aliases follow partner policy). Partner banks get structured mule clues as advisory tips. Not in-session behavioral biometrics — destination account from the victim artifact. We never freeze accounts.
Malaysia lost RM2.97 billion to online fraud in 2025 (Home Ministry), with 66,204 cases reported (PDRM) — nearly double the prior year. The current response is fragmented and manual — victims report hours after the scam, mule accounts are drained within minutes, and raw evidence sharing across banks risks PDPA violations.
The Waspada SDK runs inside the host banking app or the public triage sandbox. No receipt or screenshot ever uploads to our servers. All extraction happens on-device before transmission.
Handles major Malaysian banking receipts, screenshots, and multi-page PDFs — including encrypted files. Extraction is CI-gated on clean and noisy labeled corpora; we publish characterized misses, not vanity 100%s.
Three-tier extraction identifies mule accounts, BIC codes, scammer aliases, and malicious URLs. Runs natively on Apple and Android devices.
NRIC, phone numbers, and emails stripped before anything leaves your device. Payload cryptographically signed for tamper-proof delivery. Works offline after models are cached on first load.
Public sandbox triage stays on the phone (draft report, no auto bank feed). Partner-integrated apps can send scrubbed indicators to the fraud team — advisory only, bank decides. WhatsApp is not a live public ingest door.
Privacy compliance, extraction accuracy, and zero-friction integration with existing bank infrastructure — by design, not afterthought.
NRIC, phone, and email are stripped on-device and re-verified at the gateway. Receipt images are not uploaded for cloud inference. Only scrubbed indicators leave the device; scammer aliases egress only under partner policy.
Measured on labeled Malay receipt corpora — clean CI gate plus a noisy soft gate with published failure modes (including wrong-account vs miss). Parsing stays on-device; alerts stay advisory so banks own any freeze.
Every alert is cryptographically signed with per-partner key isolation. Replay attacks and payload tampering are structurally prevented. Image authenticity is not assumed from a single receipt — corroboration belongs with the bank.
Live streaming alerts to bank fraud dashboards on the partner path. Memory fast path under a second; durable delivery ~2 seconds p99 — measured topologies, not a single blended claim.
Malicious URLs from scam messages enriched with domain registration data for attribution and takedown support.
Gateway formats advisory alerts as STIX 2.1 for analyst export / offline use. Live bank delivery is the signed FMS JSON webhook — not an automated SIEM push. Patterned indicators are suppressed while disposition is advisory.
Four endpoints. Partner-authenticated ingestion, real-time alert streaming, and health monitoring. Full API docs in the developer portal.
For bank fraud, mobile, and FMS teams evaluating an advisory evidence layer — start with contracts you can verify, then a staged integration week.