Fraud Detection Thresholds
Configure the risk score boundaries that determine transaction routing decisions. Changes take effect immediately on the live stream.
Transactions with risk score at or above this value are sent to the Review Queue for analyst inspection.
Transactions at or above this score are automatically blocked. Must be greater than the Flag threshold.
Time window (minutes) for counting repeated transactions from the same user/IP.
Number of transactions within the window that triggers a velocity fraud flag.
Minimum km between two transactions for them to be flagged as impossible travel.
Transactions from these MCC categories receive a +15 risk score penalty automatically.
AI Model Configuration
Select the Gemini models powering each agent in the fraud detection pipeline. Higher-tier models increase accuracy and cost.
Controls output randomness. Lower = more deterministic decisions.
Maximum time before an agent call is aborted and fallback rule fires.
Used by all agents. Stored in .env — never logged or transmitted.
Budget & Cost Controls
Set hard spending limits and configure the adaptive routing engine that degrades gracefully as budget runs low.
Hard cap. Agent calls are suspended when this is reached.
Soft cap — triggers alerts at 80%, 90%, 100%.
When spend counters are reset to zero.
At each utilization level, the system automatically routes to a cheaper model tier.
Send Slack / email alerts at 80%, 90%, and 100% budget utilization.
Dynatrace Integration
Connect FinSentinel to your Dynatrace environment for observability, self-healing, and SRE automation.
Your Dynatrace SaaS or Managed endpoint.
Requires: metrics.read, events.ingest, problems.read
Allow the monitor to auto-remediate problems detected by Dynatrace.
Stream fraud KPIs (risk scores, decision counts) as custom metrics.
Connect Gemini agents to Dynatrace via the Model Context Protocol server.
How often the self-monitor queries Dynatrace for open problems.
Notification Rules
Configure alert channels and the conditions that trigger them. Alerts fire in real-time as transactions are processed.
Agent Pipeline Configuration
Control which agents are active in the fraud detection pipeline and tune their individual behavior.
How frequently the system processes a new simulated transaction. Lower = more Gemini calls = higher cost.