Amplinth Meridian is an AI client-lifecycle platform for relationship and commercial banks. KYC/AML, identity verification, Companies House lookup, document extraction and credit assessment run in parallel across a governed seven-stage pipeline — with a human decision at every gate and an append-only audit trail behind every step.
KYC, credit and compliance run one after another, your RMs lose most of their week to admin, and the evidence a regulator wants lives in five different systems.
The disciplined Amplinth loop, tuned to client lifecycle management — AI gathers and reconciles the evidence; your bankers and analysts make every decision.
Companies House, an Onfido identity client, an Equifax UK credit client, a locally-ingested sanctions/PEP database, uploaded documents & CRM lead pushes.
Five AI agents run in parallel on a Neo4j ownership graph: KYC/AML, document extraction, credit prep, compliance & RM Co-pilot. No black box.
A decision-ready package: RM brief, credit & affordability memos, computed risk levels, gate status and a streaming Co-pilot you can ask anything.
Humans act at every gate — analysts approve or override screening, branch managers record a two-step-confirmed decision with conditions.
Every action is written to an append-only audit log with jurisdiction-correct retention — full FCA/PRA or FINMA traceability.
A deterministic stage machine drives every client from expression of interest to activation. No client advances until the compliance gates for the next stage pass — and every gate clearance is a documented human decision.
Intake captures entity name, type, sector, product, referral and RM assignment as a structured record, with duplicate-prevention built in. Upstream CRM and origination systems can push leads in machine-to-machine via a per-tenant API key.
The RM Co-pilot aggregates a full client snapshot — entities, owners, property, latest credit — into a pre-meeting brief and meeting-prep narrative, so the first conversation is informed, not exploratory.
A risk-aware document checklist requests proof of ID/address, audited accounts and certificates of incorporation; a two-model vision-plus-text pipeline extracts and cross-checks fields with per-field confidence. A secure client portal collects documents over time-limited, OTP-verified upload links.
A centralised CDD workbench queue lets analysts claim, escalate, request information, return or approve each case, with the AI rationale alongside to confirm or override. High-risk and complex cases route to an escalation lead.
A business credit check, risk-assessment engine and credit/affordability calculators compute lending metrics — LTV, ICR, DSCR for businesses, income multiples for individuals — and draft a credit memo for analyst review.
The branch manager records the lending decision with a two-step confirmation and attaches lending conditions; condition closure is tracked to fulfilment. Branch autonomy is preserved by design.
Once all gates clear, an onboarding agent builds the account-setup checklist — account opening, online banking, payment authorities and mandates — to close the loop.
The moment a client is created, five back-end agents launch together on a worker queue — so KYC, credit and compliance stop waiting on each other. Each is governed by input/output guardrails: PII masking, prompt-injection detection, schema validation, hallucination and toxicity checks.
Identifies ultimate beneficial owners (≥25%) from Companies House PSC data, verifies identities, and screens for sanctions and politically-exposed persons against a self-hosted sanctions/PEP database using Jaro-Winkler fuzzy matching — no dependence on a paid screening API, and PII stays in the bank's environment.
Pulls the registered profile, officers and persons-with-significant-control live from Companies House; retrieves filing history and the certificate of incorporation; and attaches registry documents straight to the client file for evidencing.
A two-model pipeline — a vision LLM extracts structured fields, a text LLM validates them — returns extracted data with per-field confidence and notes, ready for reviewer accept-or-correct against the source.
Runs the business credit check, computes risk and lending metrics, assesses individual affordability, and drafts a credit memo from the assembled assessment for analyst review.
Re-evaluates the gates for the client's current and next stage — mandatory KYC docs validated, identity verified, PEP/sanctions clear, CDD approved, credit complete — against a jurisdiction-aware AML matrix that adapts sanctions lists, retention and source-of-wealth rules per tenant.
Builds the relationship-manager brief, and a streaming natural-language chat answers questions over the client's data — with answers saveable as case notes.
The platform spine — the gated pipeline, parallel agents, live Companies House integration, self-hosted sanctions screening, risk + credit + affordability engines and immutable audit — is deployed and running today. The business-impact figures below are modelled pilot projections, not yet realized client results.
A complete, gated onboarding spine — pipeline, parallel agents, KYC, credit and decisioning — running multi-tenant in the cloud.
Roadmap, not claimed today: adverse-media screening and scheduled sanctions ingestion are built and near wiring; transaction monitoring, SAR/STR workflow, continuous re-screening, a native iPad client and the Swiss Zefix registry are specified for the roadmap. We label maturity honestly so you can plan with confidence.
Connect Meridian to Companies House, your identity and credit providers, sanctions sources and CRM. Configure the tenant — jurisdiction, branding, locale, currency, sanctions lists and risk rules, no code required. Operate the gated pipeline in production, where a human stays accountable at every gate and one codebase serves UK and Swiss regimes across four locales.
Meridian automates the evidence-gathering, never the judgement. The org-unit branch model, two-step human decision and "AI subordinate to human" are enforced in the gate machine — you speed onboarding without changing your relationship-led operating model.
Input/output guardrails, an append-only audit log tied to FCA/PRA and FINMA requirements, and an override trail on every AI output. This is Amplinth's edge over generic decision-intelligence tools — decisions you can defend.
One codebase serves UK and Swiss regimes via tenant config — sanctions lists, retention, source-of-wealth, PEP handling, currency and four locales — rather than separate forks. A self-hosted sanctions screen and a Neo4j ownership graph round out the spine.
Book a 30-minute demo on your toughest high-risk case — we'll show you the parallel agents, the gated pipeline, and the audit trail that proves every decision.