From cc86b144ac45dc23da9734fa846c94455cbbfcc8 Mon Sep 17 00:00:00 2001
From: patrick-motsch <p.motsch@valueon.ch>
Date: Sun, 15 Feb 2026 22:52:02 +0100
Subject: [PATCH] fix(teamsbot): filter bot's own captions from AI trigger,
 mark bot responses in context to prevent repetition

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 modules/features/teamsbot/service.py | 44 ++++++++++++++++++++++++++--
 1 file changed, 42 insertions(+), 2 deletions(-)

diff --git a/modules/features/teamsbot/service.py b/modules/features/teamsbot/service.py
index 47a036fc..ad52e0c5 100644
--- a/modules/features/teamsbot/service.py
+++ b/modules/features/teamsbot/service.py
@@ -307,6 +307,11 @@ class TeamsbotService:
         if not text:
             return
 
+        # Filter out the bot's own speech from AI triggering.
+        # The bot hears itself via captions — these should be stored in the
+        # transcript for the record, but must NOT trigger AI analysis (feedback loop).
+        isBotSpeaker = self._isBotSpeaker(speaker)
+
         # Store transcript segment
         transcriptData = TeamsbotTranscript(
             sessionId=sessionId,
@@ -346,6 +351,11 @@ class TeamsbotService:
             count = session.get("transcriptSegmentCount", 0) + 1
             interface.updateSession(sessionId, {"transcriptSegmentCount": count})
 
+        # Skip AI analysis for bot's own speech (prevents feedback loop)
+        if isBotSpeaker:
+            logger.debug(f"Session {sessionId}: Skipping AI trigger for bot's own speech: [{speaker}] {text[:60]}...")
+            return
+
         # Check if AI analysis should be triggered (only for final transcripts)
         if not isFinal:
             return
@@ -364,6 +374,32 @@ class TeamsbotService:
         logger.info(f"Session {sessionId}: Triggering AI analysis (buffer: {len(self._contextBuffer)} segments)")
         await self._analyzeAndRespond(sessionId, interface, voiceInterface, websocket, createdTranscript)
 
+    def _isBotSpeaker(self, speaker: str) -> bool:
+        """Check if a transcript speaker is the bot itself.
+        
+        Teams captions show the bot as e.g. "Shelly Miller (Unverified)" or 
+        "Nyla Larsson" depending on auth/anonymous join. We match against:
+        - The configured bot name (e.g. "Shelly Miller")
+        - The bot account display name if authenticated
+        """
+        if not speaker:
+            return False
+        
+        speakerLower = speaker.lower().strip()
+        
+        # Match against configured bot name
+        botName = self.config.botName.lower().strip()
+        if botName and botName in speakerLower:
+            return True
+        
+        # Match against bot account email prefix (e.g. "nyla.larsson" from "nyla.larsson@poweron.swiss")
+        if self.config.botAccountEmail:
+            emailPrefix = self.config.botAccountEmail.split("@")[0].lower().replace(".", " ")
+            if emailPrefix in speakerLower:
+                return True
+        
+        return False
+
     def _shouldTriggerAnalysis(self, transcriptText: str) -> bool:
         """
         Decide whether to trigger AI analysis based on the latest transcript.
@@ -405,12 +441,16 @@ class TeamsbotService:
         """Run SPEECH_TEAMS AI analysis and respond if needed."""
         self._lastAiCallTime = time.time()
 
-        # Build transcript context from buffer
+        # Build transcript context from buffer.
+        # Mark bot's own utterances so the AI knows what it already said.
         contextLines = []
         for segment in self._contextBuffer:
             speaker = segment.get("speaker", "Unknown")
             text = segment.get("text", "")
-            contextLines.append(f"[{speaker}]: {text}")
+            if self._isBotSpeaker(speaker):
+                contextLines.append(f"[YOU ({self.config.botName})]: {text}")
+            else:
+                contextLines.append(f"[{speaker}]: {text}")
         
         transcriptContext = f"BOT_NAME:{self.config.botName}\n" + "\n".join(contextLines)