ESCALATION REQUEST: Latency Critical Architecture (EN 81-28 P100 Protocol) - Engineering Input Needed
Hi Team,
I received an automated/generic response to my previous inquiry, but this is a critical architectural blocker that requires input from a Solutions Engineer.
Context:
I am building an elevator emergency dispatch system (High-Risk AI under EU AI Act).
The system MUST comply with the P100 protocol:
- Incoming call sends DTMF ‘D’.
- System must respond with DTMF ‘A’ within <1000ms.
The standard LLM pipeline (STT → LLM → Tool Call) is too slow (~1500ms+), causing hardware timeouts.
I need a definitive technical answer on two specific workarounds:
OPTION A (Static Audio Bypass):
If I upload a Custom Audio File (if thats even possible somehow) as the first_message (containing a 200ms DTMF ‘A’ tone followed by silence and the greeting), does the Retell engine play this immediately upon SIP connection (0ms latency), or is there still a processing delay/buffer before audio starts?
OPTION B (SIP-Level Logic):
Does Retell support any “Pre-LLM” logic hooks? Can I configure the agent to listen for a specific DTMF tone before sending the stream to the LLM, and trigger a raw PressDigit action immediately?
Please route this to a technical specialist. We cannot proceed with the implementation without confirming we can beat the 1000ms latency requirement.
Thanks,
Rogier Helvensteijn
Reflow Automations