Sameer Bhalerao
All projects

AML/KYC compliance intelligence · Luxembourg

ComplyLens

Compliance intelligence that makes CSSF-inspection-ready documentation auditable — reasoning visible, evidence cited, output defensible.

Demo live — complylens.vercel.app · pre-seed exploration

The problem

Luxembourg AML/KYC compliance teams produce manual review documentation that is difficult to make "inspection-ready" for CSSF audits. Analysts classify clients, write periodic review memos, and disposition alerts — all in documents with no structured evidence chain. In 2024, CSSF handed down €4.7M in fines (up 52%). The problem isn't whether a review happened; it's whether the reasoning was auditable.

How it works

  • Three product pillars sit on a shared FastAPI backend (Railway) + PostgreSQL database: RiskRationale, ReviewMemo, and AlertDesk.
  • RiskRationale classifies client risk and generates structured evidence citations mapped to CSSF risk indicators — not a black-box score, but a reasoned classification you can defend in an inspection.
  • ReviewMemo generates 11-section periodic review memos with a companion CSSF audit pack (ZIP export) — Claude LLM drafts the entire structure; the compliance officer reviews and signs.
  • AlertDesk handles sanctions and PEP alert dispositions with a four-eyes workflow — Claude LLM drafts the disposition; a second reviewer approves before the record is locked.
  • A full audit log records every AI determination. Analysts can correct AI output through a structured domain-correction loop that feeds back into the system. A codified regulatory_framework module maps every evidence citation to a specific CSSF indicator — nothing is asserted without a regulatory anchor.
  • A password-gated live demo (Meridian Holdings synthetic scenario) is accessible to prospects at complylens.vercel.app. A demo-reset endpoint makes demos repeatable. Regulatory timing: AMLR applies July 2027, creating a first-mover window for tools built on auditable LLM reasoning.

Architecture

Client interface

Next.js 15 — Vercel

password-gated demo · complylens.vercel.app

API requests

FastAPI Backend — Railway + PostgreSQL

RiskRationaleReviewMemoAlertDeskFour-eyes workflowDemo reset endpoint
AI reasoning

AI + output layer

Claude

structured reasoning · CSSF-mapped evidence citations

WeasyPrint

11-section PDF + CSSF audit ZIP export

Context

CSSF 2024: €4.7M fines (+52%)AMLR applies July 2027Luxembourg ManCos + AIFMsMeridian Holdings demo scenario

Claude LLM drafts structured reasoning with CSSF-mapped evidence citations; humans review and sign; every determination is auditable.

The journey

The compliance analyst's job isn't to find the answer — it's to build the record proving the answer is right. ComplyLens does the drafting so they can focus on the judgement.

01

The analyst opens their review queue — everything assigned, nothing in email

Periodic reviews, alert dispositions, pending sign-offs — all in one workspace, all assigned, status visible at a glance. No chasing shared drives. No losing track of where a client sat in the process.

Review queue — assigned, in-review, and signed in one workspace

02

They select a client — Claude generates the complete memo in seconds

11 sections, a risk classification, CSSF-mapped evidence citations, and a structured determination — drafted instantly. The analyst's job shifts from writing to review. They add judgement, catch errors, and sign. The record is locked and audit-ready.

Claude drafts the full memo — analyst reviews, adds judgement, signs

Highlights

  • 3 product pillars: RiskRationale · ReviewMemo · AlertDesk
  • CSSF audit pack: 11-section memo + ZIP export
  • Four-eyes workflow on all alert dispositions
  • Full audit log + analyst correction feedback loop
  • regulatory_framework module — every citation mapped to a CSSF indicator
  • Full demo seeded — repeatable via demo-reset endpoint
  • AMLR July 2027 regulatory timing advantage

Stack

FastAPINext.js 15PostgreSQLRailwayVercelClaudeWeasyPrintCeleryRedisZustandTailwind 4

Why it matters

Compliance intelligence at the intersection of LLM product design and regulation — where the design constraint is not just "does the LLM output work" but "is the reasoning auditable by a regulator." Built MVP in 2 days to demo-ready state.