Switch transcription to Whisper large v3

k8s/configmap.yaml

@@ -11,11 +11,11 @@ data:
   LLM_BASE_URL: "http://10.2.3.5:8002"
   LLM_MODEL: "qwen2.5-14b"
   LLM_TIMEOUT: "5m"
-  # Voxtral Small is the only transcription provider. It is exposed on the AI
-  # server through an OpenAI-compatible /v1/audio/transcriptions endpoint.
-  VOXTRAL_BASE_URL: "http://10.2.3.5:8004"
-  VOXTRAL_MODEL: "mistralai/Voxtral-Small-24B-2507"
-  VOXTRAL_TIMEOUT: "30m"
+  # Whisper Large v3 is exposed on the AI server through an OpenAI-compatible
+  # /v1/audio/transcriptions endpoint.
+  AUDIO_TRANSCRIPTION_BASE_URL: "http://10.2.3.5:8004"
+  AUDIO_TRANSCRIPTION_MODEL: "openai/whisper-large-v3"
+  AUDIO_TRANSCRIPTION_TIMEOUT: "30m"
   AI_STATS_SIDECAR_URL: "http://10.2.3.5:9090"
   AI_STATS_TIMEOUT: "8s"
   WORKER_POLL_INTERVAL: "2s"

k8s/secrets.yaml

@@ -18,5 +18,5 @@ type: Opaque
 stringData:
   DATABASE_URL: "postgres://ai_service:ai_service@postgres:5432/ai_service?sslmode=disable"
   LLM_API_KEY: "sk-111f838ccec43406e078cd9094b6797307cb895236179f32"
-  VOXTRAL_API_KEY: "sk-111f838ccec43406e078cd9094b6797307cb895236179f32"
+  AUDIO_TRANSCRIPTION_API_KEY: "sk-111f838ccec43406e078cd9094b6797307cb895236179f32"
   AI_SERVICE_TOKEN: "d18bcacf9e02bae1806ee6b6eeda62b95be6a915c0a22936d9a700128b275442"

k8s/worker-deployment.yaml

@@ -98,7 +98,7 @@ spec:
             - name: WORKER_TASK_TYPES
               value: "transcription"
             - name: WORKER_MODEL_PROFILES
-              value: "voxtral-small"
+              value: "whisper-large-v3"
             - name: WORKER_CLAIM_LIMIT
               value: "2"
             - name: WORKER_LEASE_TIMEOUT