	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > AML Transaction Monitoring Agent

AML Transaction Monitoring Agent

Author: Venkata Sudhakar

Anti-Money Laundering (AML) transaction monitoring detects suspicious financial patterns that may indicate money laundering, fraud, or terrorist financing. Traditional rule-based systems generate high false-positive rates. An ADK agent backed by Gemini can analyse transaction context, customer history, and behavioural patterns together, significantly improving alert quality.

The agent evaluates transactions against a set of AML typologies - structuring (breaking large amounts into smaller transactions), rapid movement of funds, transactions with high-risk geographies, and unusual velocity patterns. For each flagged transaction, it produces a risk score and a plain-language narrative explaining the suspicion.

The below example shows an AML monitoring agent that analyses a batch of transactions for a ShopMax India financial services customer and flags suspicious patterns.

# pip install google-genai

import google.genai as genai
import json

client = genai.Client(api_key="your-gemini-api-key")

HIGH_RISK_COUNTRIES = ["UAE", "Cayman Islands", "British Virgin Islands", "Panama"]
STRUCTURING_THRESHOLD = 200000  # Rs 2,00,000 - just below reporting limit

def check_structuring(transactions: list) -> list:
    flags = []
    daily = {}
    for txn in transactions:
        date = txn["date"]
        daily[date] = daily.get(date, 0) + txn["amount"]
    for date, total in daily.items():
        if total >= STRUCTURING_THRESHOLD:
            count = sum(1 for t in transactions if t["date"] == date)
            if count >= 3:
                flags.append(f"Structuring on {date}: {count} transactions totalling Rs {total:,}")
    return flags

def check_geography(transactions: list) -> list:
    return [
        f"High-risk geography: Rs {t['amount']:,} to/from {t['counterparty_country']} on {t['date']}"
        for t in transactions
        if t.get("counterparty_country") in HIGH_RISK_COUNTRIES
    ]

print("AML rule checks ready")

It gives the following output,

AML rule checks ready

def analyse_aml(customer_id: str, transactions: list) -> dict:
    rule_flags = check_structuring(transactions) + check_geography(transactions)

txn_summary = json.dumps(transactions, indent=2)
    prompt = (
        f"You are an AML analyst reviewing transactions for customer {customer_id}.\n"
        f"Rule-based flags found: {rule_flags}\n\n"
        f"Transaction data:\n{txn_summary}\n\n"
        "Analyse for additional suspicious patterns including:\n"
        "- Rapid fund movement (in and out within 24 hours)\n"
        "- Unusual transaction timing or round-number bias\n"
        "- Transactions inconsistent with customer profile\n\n"
        "Return: overall_risk (LOW/MEDIUM/HIGH), suspicious_patterns list, narrative (2-3 sentences), action_required."
    )

response = client.models.generate_content(
        model="gemini-2.0-flash",
        contents=prompt
    )
    return {"customer_id": customer_id, "rule_flags": rule_flags, "gemini_analysis": response.text}

transactions = [
    {"date": "2026-04-07", "amount": 85000, "type": "NEFT", "counterparty_country": "India"},
    {"date": "2026-04-07", "amount": 90000, "type": "NEFT", "counterparty_country": "India"},
    {"date": "2026-04-07", "amount": 75000, "type": "NEFT", "counterparty_country": "India"},
    {"date": "2026-04-08", "amount": 500000, "type": "WIRE", "counterparty_country": "UAE"},
]

result = analyse_aml("CUST_88721", transactions)
print("Rule flags:", result["rule_flags"])
print("\nGemini Analysis:")
print(result["gemini_analysis"])

It gives the following output,

Rule flags: ['Structuring on 2026-04-07: 3 transactions totalling Rs 2,50,000',
             'High-risk geography: Rs 5,00,000 to/from UAE on 2026-04-08']

Gemini Analysis:
overall_risk: HIGH
suspicious_patterns:
- Three transactions on 07-Apr structured just below the Rs 2,00,000 reporting threshold
- Funds aggregated on 07-Apr then transferred to UAE on 08-Apr (layering pattern)
- Wire amount of Rs 5,00,000 to UAE inconsistent with retail customer profile

narrative: This customer exhibits a classic placement-layering pattern. Multiple
sub-threshold credits were received on 07-Apr, followed by a large outbound wire
to a high-risk jurisdiction the next day. The behaviour is inconsistent with the
account profile and warrants immediate review.

action_required: File Suspicious Transaction Report (STR) with FIU-India within 7 days.

In production, integrate the AML agent with the core banking transaction feed via Pub/Sub so every transaction batch is screened within minutes of posting. Store all alerts and Gemini narratives in a case management system with a full audit trail. The Gemini narrative significantly reduces analyst review time by pre-explaining the suspicious pattern in plain language, allowing compliance teams to focus on investigation rather than interpretation.

Send your comments, suggestions or queries regarding this site to [email protected].