A real, browser-based AML platform aligned with Bangladesh Bank Circular 26/2024. Detection runs against your own transaction file. Alerts surface with explainable reasons. STR drafts arrive ready for CAMLCO review.
Secure briefing intake protocol
8
Bangladesh-tuned detection rules in production. Structuring, layering, fan-in, fan-out, dormant spike, more.
engine/app/core/detection/rules
4
Banks already on the cross-bank intelligence layer. Each new tenant strengthens every other tenant's signal.
kestrel-engine · matches table
BDT
Native pricing, native channels, native typologies. NPSB / BEFTN / RTGS / bKash / Nagad / Rocket out of the box.
Bangladesh-only by design
4 wks
From signed order to first STR drafted in production. No multi-quarter integration project.
Standard onboarding plan
Three modules. Each replaces a manual workflow that today eats analyst hours and lets real signal slip through. All three are live in the platform now — not on a roadmap.
Eight production rules tuned for Bangladesh: rapid cashout, fan-in burst, fan-out burst, structuring, layering, first-time high value, dormant spike, proximity to flagged. Run nightly across your portfolio or call the scoring API per transaction.
Sample hits
Claude Sonnet 4.6, routed via OpenRouter, transforms rule hits + entity context into the kind of paragraph an analyst would write at the start of an STR. Names redacted at the prompt boundary; full audit trail on every model call.
Sample hits
One click on a flagged alert produces a populated draft STR — subjects resolved, accounts attached, narrative pre-written, typology pre-tagged. CAMLCO reviews, edits, and submits. goAML XML is the export format Bangladesh Bank already accepts.
Sample hits
Scam money does not stay inside one bank. It moves across six in twelve minutes. Kestrel is the only AML surface in Bangladesh that resolves entities across every participating bank — accounts, phones, wallets, NIDs, devices — and surfaces a peer warning the moment a counterparty already burned another bank touches yours.
Your own book stays your own. Peer bank names are anonymised before the data leaves the engine. Your CAMLCO desk sees a flag, a confidence, and a typology — never a competitor's book. The persona invariants are enforced in the service layer and backed by unit tests + Postgres RLS.
The 2024 AML/CFT modernisation cycle pushes scheduled banks toward risk-based monitoring, faster STR cycles, and higher-fidelity analyst evidence. Kestrel was built around those obligations from day one — not retrofitted to them.
Reference · BB Circular 26/2024 · AML/CFT instructions for scheduled banks
Risk-based transaction monitoring with documented rules, thresholds, and reasoning. Kestrel ships 8 rules with YAML definitions, weighted scoring, and per-hit traces.
Timely STR submission to BFIU in goAML XML. Kestrel imports goAML XML from existing pipelines and exports STR / SAR / CTR / TBML / IER / 5 more variants in goAML-compliant XML.
Auditable analyst workflow with immutable evidence trail. Kestrel writes every action to an append-only audit log; STR drafts, edits, submissions, and dispositions are all timestamped and attributed.
Cross-institutional intelligence sharing where permitted. Kestrel resolves entities across participating banks with peer-anonymised views and BFIU-only full-data access.
Annual licence per institution. No per-transaction metering on Starter or Professional. Enterprise meters the real-time scoring API only. Procurement-ready quotation issued within 5 working days of a signed NDA.
Tk 60 lakh
annual · 1 institution
The smallest banks. Daily batch monitoring, AI-assisted alert review, goAML round-trip.
Tk 1.5 crore
annual · 1 institution
Mid-sized banks running an active CAMLCO desk. Cross-bank intelligence layer included.
Tk 4 crore
annual · 1 institution
The largest banks and the ones moving toward real-time core-banking integration.
CSV, XLSX, or goAML XML. NPSB, BEFTN, RTGS, MFS, cash, cheque, card, wire — whatever your core system speaks. The same parser feeds the nightly scan and the upload-a-file path.
┼Eight detection rules run across every account in the batch. Entities resolve across past and present scans. The scorer produces a 0-100 risk number with per-rule contributions you can defend in writing.
┼Anything past the action threshold lands on the analyst queue with severity, evidence, and a Claude-written narrative. Cross-bank matches show as an extra flag; the dossier is one click away.
┼From the alert, your CAMLCO triggers a draft STR — subjects resolved, accounts attached, typology pre-tagged, narrative pre-written. Review, edit, submit. goAML XML on the way out.
┼Self-serve signup spins up an isolated bank tenant pre-loaded with a synthetic Bangladesh dataset — exercise detection, alerts, draft STRs, cross-bank intelligence on day one. Or file a briefing intake and we walk through it with you on a 30-minute call.