System Foundations
ANET Overview and Operating Model
Non-Emergency Scope and Boundaries
ANET is intentionally scoped to non-emergency call handling. Emergency situations require immediate human response. Uncertainty should favor escalatio...
What ANET is Designed to Do
ANET is a conversational AI system designed to handle non-emergency public safety calls. It is built to: Understand caller intent using natural langua...
Core System Architecture
Every ANET call involves three coordinated systems. STT: Converts caller speech into transcript text Detects pauses and end-of-turn signals Provides c...
Conversational Model vs Menu-Based Systems
ANET operates as a conversational system, not a traditional phone tree. Menu-based systems: Require keypad input Follow rigid branches Fail when calle...
Intent Identification Model
Each call must resolve to an intent. An intent represents the caller’s goal. Examples: Noise complaint Parking issue Request officer Animal control Sp...
Safety-First Design Philosophy
ANET is intentionally conservative. Throughout the call, ANET evaluates for: Explicit human request Mentions of violence or danger Unsafe ambiguity Sy...
Guardrails and Operational Limits
ANET includes built-in guardrails to prevent drift or instability. Examples include: Time limits on calls Maximum language switches Confidence thresho...
What Supervisors and Administrators Should Understand
Supervisors primarily: Review completed calls Monitor transfer behavior Identify trends Validate safety escalation patterns Oversight tools support re...
System Design Principles
ANET is built on the following principles: Conversational, not menu-driven Safety-biased escalation Configurable but controlled Transparent through st...
Limitations
ANET does not: Replace emergency dispatch Guarantee perfect intent detection Eliminate transfers Operate independently of telephony configuration Over...
Key Takeaways
ANET handles non-emergency calls using conversational AI. Safety overrides always take priority. The system adapts to callers rather than forcing menu...
Safety Model and Escalation Logic
Safety as a Core Control Layer
Safety is not a feature layered on top of ANET. It is a continuous control layer active throughout every call. Safety logic: Operates independently of...
Escalation Triggers
Escalation may occur when: A caller explicitly asks to speak to a person Language indicates violence, weapons, or immediate danger Ambiguity creates s...
Types of Escalation
Triggered when: The caller explicitly asks for a dispatcher or a person Behavior: Immediate transfer Intent flow bypassed Triggered by: Dangerous lang...
Over-Transfer Philosophy
ANET is intentionally biased toward escalation when uncertainty exists. Design stance: It is acceptable to transfer a call that might have been automa...
How Escalations Appear in Review Tools
In oversight tools, supervisors may see: Transfer reason codes Timeline markers Summary annotations Safety classification tags Escalations are visible...
Supervisor Review Guidance
Supervisors should: Review repeated safety transfers within a short window Evaluate unknown or agent_failure cases Confirm telephony health during gua...
Administrative Considerations
Administrators should understand: Safety overrides cannot be configured away Time limits and guardrails protect system stability Over-optimization for...
Limitations of Automation in Risk Detection
ANET: Relies on language signals Does not replace human judgment Cannot interpret non-verbal cues Operates within defined detection thresholds Human o...
Key Takeaways
Safety is a continuous control layer. Escalation overrides intent routing. Transfers are often protective, not failures. Guardrails prevent drift and ...