ES
joseph
Say hi ↗
PARSE • SCORE • PROTECT • PARSE • SCORE • PROTECT •
case 02 / 05

Sentinel Guard 

AI-driven security bridge for the VUCEM customs platform that detects anomalies in XML declarations.

Year
2024
Role
Backend · Integrations
Status
Private · company network only
( architecture )

How it's built

PythonAWS BedrockPostgreSQLC#SQSEventBridge
xmlrisksoap▭Customs ERPclientλSOAP BridgeC# service◉S3 Raw XMLbucket≡EventBridgerulesλXML Parserλ python≡SQSscoring queue✦Anomaly Scorerλ + Bedrock◉PostgreSQLRDS≡SNS Alertstopic↗VUCEMexternal
Hover a service to trace its connections
( the problem )

Customs declarations were reviewed after submission, so malformed or suspicious XML was only caught once it had already cost time and fines.

( the build )

Every declaration is intercepted, parsed and scored by a Bedrock model before it reaches VUCEM. Risky ones are held and flagged to auditors in real time.

94%
anomalies caught pre-submit
<2s
scoring per declaration
0
XML rejections last quarter
( request flow )
  1. 01The C# SOAP bridge drops every outgoing XML declaration into S3.
  2. 02EventBridge triggers a Python parser that normalises it into structured fields.
  3. 03SQS buffers scoring jobs; a Bedrock-backed Lambda flags anomalies.
  4. 04Results land in PostgreSQL and SNS alerts the audit team before submission.
sentinel.console● live
94%
anomalies caught pre-submit
<2s
scoring per declaration
0
XML rejections last quarter
handler.py
1score = bedrock.invoke_model(
2 modelId=ANOMALY_MODEL,
3 body=declaration.to_prompt(),
4)
5if score.risk > 0.82: hold(decl)
6▍
main ✓Python · UTF-8 · Ln 6
Next case · 03
Lumina Vision CRM →