Merge branch 'develop'

README.md

@@ -472,8 +472,8 @@ Conversation options are represented as features. They can be configured from Ap
 
 | Name | Description | Type | Default |
 |-|-|-|-|
-| `answer_hard_timeout_sec` | The hard timeout for the bot answer in secs. | `int` | 60 |
-| `answer_soft_timeout_sec` | The soft timeout for the bot answer in secs. | `int` | 30 |
+| `answer_hard_timeout_sec` | Time waiting the LLM before aborting the answer with an error message. | `int` | 40 |
+| `answer_soft_timeout_sec` | Time waiting the LLM before sending a waiting message. | `int` | 15 |
 | `callback_timeout_hour` | The timeout for a callback in hours. Set 0 to disable. | `int` | 3 |
 | `phone_silence_timeout_sec` | Amount of silence in secs to trigger a warning message from the assistant. | `int` | 20 |
 | `recognition_retry_max` | TThe maximum number of retries for voice recognition. Minimum of 1. | `int` | 3 |

app/helpers/call_llm.py

@@ -201,8 +201,8 @@ async def _response_callback(_retry: bool = False) -> None:
                     )
                 )
 
-            # Process the response and wait for latency metrics
-            await _commit_answer(wait=False)
+            # Process the response and wait for it to be able to kill the task if needed
+            await _commit_answer(wait=True)
 
         # First call
         if len(call.messages) <= 1:
@@ -215,7 +215,7 @@ async def _response_callback(_retry: bool = False) -> None:
             )
         # User is back
         else:
-            # Welcome with the LLM, do not use the end call tool for the first message, LLM hallucinates it and this is extremely frustrating for the user
+            # Welcome with the LLM, do not use the end call tool for the first message, LLM hallucinates it and this is extremely frustrating for the user, don't wait for the response to start the VAD quickly
             await _commit_answer(
                 tool_blacklist={"end_call"},
                 wait=False,

app/helpers/features.py

@@ -17,21 +17,21 @@
 
 async def answer_hard_timeout_sec() -> int:
     """
-    The hard timeout for the bot answer in secs.
+    Time waiting the LLM before aborting the answer with an error message.
     """
     return await _default(
-        default=60,
+        default=40,
         key="answer_hard_timeout_sec",
         type_res=int,
     )
 
 
 async def answer_soft_timeout_sec() -> int:
     """
-    The soft timeout for the bot answer in secs.
+    Time waiting the LLM before sending a waiting message.
     """
     return await _default(
-        default=30,
+        default=15,
         key="answer_soft_timeout_sec",
         type_res=int,
     )

app/helpers/llm_tools.py

@@ -98,8 +98,7 @@ async def new_claim(
                         content="",
                         persona=MessagePersonaEnum.HUMAN,
                     ),
-                    # Reinsert the last message, using more will add the user message asking to create the new claim and the assistant can loop on it sometimes
-                    self.call.messages[-1],
+                    # TODO: Should it be a reminder for the last conversation subject? It would allow to keep the context of the conversation. Keeping the last message in the history is felt as weird for users (see: https://github.com/microsoft/call-center-ai/issues/397).
                 ],
             )
         )

cicd/bicep/app.bicep

@@ -949,8 +949,8 @@ resource assignmentAppConfigurationDataReader 'Microsoft.Authorization/roleAssig
 
 resource configValues 'Microsoft.AppConfiguration/configurationStores/keyValues@2023-03-01' = [
   for item in items({
-    answer_hard_timeout_sec: 180
-    answer_soft_timeout_sec: 30
+    answer_hard_timeout_sec: 40
+    answer_soft_timeout_sec: 15
     callback_timeout_hour: 3
     phone_silence_timeout_sec: 20
     recognition_retry_max: 2