AML RAG Agent (AWS Bedrock + Lambda + Knowledge Base)

A serverless AML (Anti-Money Laundering) assistant that combines:

• AWS Bedrock foundation models
• AWS Knowledge Bases for Retrieval-Augmented Generation (RAG)
• LangChain + LangGraph tool-based agent workflow
• AWS Lambda (container image deployed via SAM)
• Streamlit web interface

The system supports document retrieval, transaction risk assessment, sanctions checks, and compliance report generation.

---

Features

• Retrieve AML policies from an AWS Knowledge Base
• Tool-based transaction risk scoring
• Simulated sanctions list screening
• Generate structured AML compliance reports
• Upload documents to S3 for Knowledge Base indexing
• REST API exposed through API Gateway
• Streamlit UI for document upload and chat

---

Picture

---

Simplified Layout

src/
  agent_handler.py        # Main Lambda agent with tools + workflow
  lambda_function.py      # Simpler RAG-only Lambda handler
  requirements.txt
  Dockerfile

streamlit_app.py          # Streamlit frontend

template.yaml             # AWS SAM deployment template

---

AWS Resources Used

This project expects:

• An AWS Bedrock-enabled region
• An existing AWS Knowledge Base ID
• An S3 bucket used as the document source

Environment variables used by Lambda:

KB_ID=<knowledge-base-id>
S3_BUCKET=<bucket-id>

---

Deployment (AWS SAM)

Build the container:

sam build

Deploy:

sam deploy --guided

After deployment, SAM outputs the API endpoint:

https://<api-id>.execute-api.<region>.amazonaws.com/Prod/query

Use this endpoint in the Streamlit app configuration.

---

Lambda Container

Base image:

public.ecr.aws/lambda/python:3.10

Handler:

agent_handler.lambda_handler

Dependencies installed from:

requirements.txt

---

API Usage

Endpoint

POST /query

Request body

{
  "query": "Check a $15000 crypto transfer from IR"
}

Optional mode switch:

{
  "query": "...",
  "mode": "agent"
}

Modes:

• "agent" → tool-enabled AML agent
• "rag" → simple Knowledge Base retrieval

Example response

{
  "answer": "...",
  "mode": "agent",
  "tools_used": ["assess_transaction_risk"],
  "query": "..."
}

---

Streamlit UI (Optional)

Install minimal UI dependencies:

pip install streamlit boto3 requests

Run locally:

streamlit run streamlit_app.py

The UI allows:

• Uploading AML documents to S3
• Chatting with the assistant through the deployed API
• Viewing uploaded document history

---

Document Upload Flow

Streamlit → S3 bucket (documents/) → Knowledge Base sync → Lambda retrieval

Knowledge Base indexing may take a short time before new documents become searchable.

---

Agent Tools Implemented

Tool	Purpose
search_aml_knowledge_base	Retrieve compliance documents
assess_transaction_risk	Calculate AML risk score
check_sanctions_list	Simulated sanctions screening
get_uploaded_documents	List documents stored in S3
generate_compliance_report	Produce structured AML report

The agent workflow instructs the model to rely on tool outputs for transaction analysis.

---

Notes

• Sanctions screening logic is simplified for demonstration.
• Risk scoring logic is rule-based and not regulatory-certified.
• Knowledge Base syncing behavior depends on AWS indexing configuration.

---