♻️(summary) clean up code and unify logging in preparation for testing

Refactor the summary service to better separate concerns, making components
easier to isolate and test. Unify logging logic to ensure consistent
behavior and reduce duplication across the service layer. These changes
set up the codebase for granular testing.

CHANGELOG.md

@@ -11,6 +11,7 @@ and this project adheres to
 ### Added
 
 - 👷(docker) add arm64 platform support for image builds
+- ✨(summary) add localization support for transcription context text
 
 ### Changed
 

gitlint/gitlint_emoji.py

@@ -2,6 +2,7 @@
 Gitlint extra rule to validate that the message title is of the form
 "<gitmoji>(<scope>) <subject>"
 """
+
 from __future__ import unicode_literals
 
 import re

src/backend/core/recording/event/notification.py

@@ -167,6 +167,7 @@ class NotificationService:
                 owner_access.user.timezone
             ).strftime("%H:%M"),
             "download_link": f"{get_recording_download_base_url()}/{recording.id}",
+            "context_language": owner_access.user.language,
         }
 
         headers = {

src/summary/summary/api/route/tasks.py

@@ -15,8 +15,8 @@ from summary.core.config import get_settings
 settings = get_settings()
 
 
-class TaskCreation(BaseModel):
-    """Task data."""
+class TranscribeSummarizeTaskCreation(BaseModel):
+    """Transcription and summarization parameters."""
 
     owner_id: str
     filename: str
@@ -28,6 +28,7 @@ class TaskCreation(BaseModel):
     recording_time: Optional[str]
     language: Optional[str]
     download_link: Optional[str]
+    context_language: Optional[str] = None
 
     @field_validator("language")
     @classmethod
@@ -45,8 +46,8 @@ router = APIRouter(prefix="/tasks")
 
 
 @router.post("/")
-async def create_task(request: TaskCreation):
-    """Create a task."""
+async def create_transcribe_summarize_task(request: TranscribeSummarizeTaskCreation):
+    """Create a transcription and summarization task."""
     task = process_audio_transcribe_summarize_v2.apply_async(
         args=[
             request.owner_id,
@@ -59,6 +60,7 @@ async def create_task(request: TaskCreation):
             request.recording_time,
             request.language,
             request.download_link,
+            request.context_language,
         ],
         queue=settings.transcribe_queue,
     )

src/summary/summary/core/analytics.py

@@ -112,19 +112,16 @@ class MetadataManager:
         if self._is_disabled or self.has_task_id(task_id):
             return
 
-        initial_metadata = {
-            "start_time": time.time(),
-            "asr_model": settings.whisperx_asr_model,
-            "retries": 0,
-        }
-
         _, filename, email, _, received_at, *_ = task_args
 
+        start_time = time.time()
         initial_metadata = {
-            **initial_metadata,
+            "start_time": start_time,
+            "asr_model": settings.whisperx_asr_model,
+            "retries": 0,
             "filename": filename,
             "email": email,
-            "queuing_time": round(initial_metadata["start_time"] - received_at, 2),
+            "queuing_time": round(start_time - received_at, 2),
         }
 
         self._save_metadata(task_id, initial_metadata)

src/summary/summary/core/celery_worker.py

@@ -10,14 +10,13 @@ import openai
 import sentry_sdk
 from celery import Celery, signals
 from celery.utils.log import get_task_logger
-from requests import Session, exceptions
-from requests.adapters import HTTPAdapter
-from urllib3.util import Retry
+from requests import exceptions
 
 from summary.core.analytics import MetadataManager, get_analytics
 from summary.core.config import get_settings
 from summary.core.file_service import FileService, FileServiceException
 from summary.core.llm_service import LLMException, LLMObservability, LLMService
+from summary.core.locales import get_locale
 from summary.core.prompt import (
     FORMAT_NEXT_STEPS,
     FORMAT_PLAN,
@@ -29,6 +28,7 @@ from summary.core.prompt import (
     PROMPT_USER_PART,
 )
 from summary.core.transcript_formatter import TranscriptFormatter
+from summary.core.webhook_service import submit_content
 
 settings = get_settings()
 analytics = get_analytics()
@@ -55,89 +55,17 @@ if settings.sentry_dsn and settings.sentry_is_enabled:
         sentry_sdk.init(dsn=settings.sentry_dsn, enable_tracing=True)
 
 
-file_service = FileService(logger=logger)
+file_service = FileService()
 
 
-def create_retry_session():
-    """Create an HTTP session configured with retry logic."""
-    session = Session()
-    retries = Retry(
-        total=settings.webhook_max_retries,
-        backoff_factor=settings.webhook_backoff_factor,
-        status_forcelist=settings.webhook_status_forcelist,
-        allowed_methods={"POST"},
-    )
-    session.mount("https://", HTTPAdapter(max_retries=retries))
-    return session
+def transcribe_audio(task_id, filename, language):
+    """Transcribe an audio file using WhisperX.
 
+    Downloads the audio from MinIO, sends it to WhisperX for transcription,
+    and tracks metadata throughout the process.
 
-def format_actions(llm_output: dict) -> str:
-    """Format the actions from the LLM output into a markdown list.
-
-    fomat:
-    - [ ] Action title Assignée à : assignee1, assignee2, Échéance : due_date
+    Returns the transcription object, or None if the file could not be retrieved.
     """
-    lines = []
-    for action in llm_output.get("actions", []):
-        title = action.get("title", "").strip()
-        assignees = ", ".join(action.get("assignees", [])) or "-"
-        due_date = action.get("due_date") or "-"
-        line = f"- [ ] {title} Assignée à : {assignees}, Échéance : {due_date}"
-        lines.append(line)
-    if lines:
-        return "### Prochaines étapes\n\n" + "\n".join(lines)
-    return ""
-
-
-def post_with_retries(url, data):
-    """Send POST request with automatic retries."""
-    session = create_retry_session()
-    session.headers.update(
-        {"Authorization": f"Bearer {settings.webhook_api_token.get_secret_value()}"}
-    )
-    try:
-        response = session.post(url, json=data)
-        response.raise_for_status()
-        return response
-    finally:
-        session.close()
-
-
-@celery.task(
-    bind=True,
-    autoretry_for=[exceptions.HTTPError],
-    max_retries=settings.celery_max_retries,
-    queue=settings.transcribe_queue,
-)
-def process_audio_transcribe_summarize_v2(
-    self,
-    owner_id: str,
-    filename: str,
-    email: str,
-    sub: str,
-    received_at: float,
-    room: Optional[str],
-    recording_date: Optional[str],
-    recording_time: Optional[str],
-    language: Optional[str],
-    download_link: Optional[str],
-):
-    """Process an audio file by transcribing it and generating a summary.
-
-    This Celery task performs the following operations:
-    1. Retrieves the audio file from MinIO storage
-    2. Transcribes the audio using WhisperX model
-    3. Sends the results via webhook
-
-    """
-    logger.info(
-        "Notification received | Owner: %s | Room: %s",
-        owner_id,
-        room,
-    )
-
-    task_id = self.request.id
-
     logger.info("Initiating WhisperX client")
     whisperx_client = openai.OpenAI(
         api_key=settings.whisperx_api_key.get_secret_value(),
@@ -145,10 +73,9 @@ def process_audio_transcribe_summarize_v2(
         max_retries=settings.whisperx_max_retries,
     )
 
+    # Transcription
     try:
-        with (
-            file_service.prepare_audio_file(filename) as (audio_file, metadata),
-        ):
+        with file_service.prepare_audio_file(filename) as (audio_file, metadata):
             metadata_manager.track(task_id, {"audio_length": metadata["duration"]})
 
             if language is None:
@@ -179,13 +106,32 @@ def process_audio_transcribe_summarize_v2(
 
     except FileServiceException:
         logger.exception("Unexpected error for filename: %s", filename)
-        return
+        return None
 
     metadata_manager.track_transcription_metadata(task_id, transcription)
+    return transcription
 
-    formatter = TranscriptFormatter()
 
-    content, title = formatter.format(
+def format_transcript(
+    transcription,
+    context_language,
+    language,
+    room,
+    recording_date,
+    recording_time,
+    download_link,
+):
+    """Format a transcription into readable content with a title.
+
+    Resolves the locale from context_language / language, then uses
+    TranscriptFormatter to produce markdown content and a title.
+
+    Returns a (content, title) tuple.
+    """
+    locale = get_locale(context_language, language)
+    formatter = TranscriptFormatter(locale)
+
+    return formatter.format(
         transcription,
         room=room,
         recording_date=recording_date,
@@ -193,34 +139,93 @@ def process_audio_transcribe_summarize_v2(
         download_link=download_link,
     )
 
-    data = {
-        "title": title,
-        "content": content,
-        "email": email,
-        "sub": sub,
-    }
 
-    logger.debug("Submitting webhook to %s", settings.webhook_url)
-    logger.debug("Request payload: %s", json.dumps(data, indent=2))
+def format_actions(llm_output: dict) -> str:
+    """Format the actions from the LLM output into a markdown list.
 
-    response = post_with_retries(settings.webhook_url, data)
+    fomat:
+    - [ ] Action title Assignée à : assignee1, assignee2, Échéance : due_date
+    """
+    lines = []
+    for action in llm_output.get("actions", []):
+        title = action.get("title", "").strip()
+        assignees = ", ".join(action.get("assignees", [])) or "-"
+        due_date = action.get("due_date") or "-"
+        line = f"- [ ] {title} Assignée à : {assignees}, Échéance : {due_date}"
+        lines.append(line)
+    if lines:
+        return "### Prochaines étapes\n\n" + "\n".join(lines)
+    return ""
 
-    try:
-        response_data = response.json()
-        document_id = response_data.get("id", "N/A")
-    except (json.JSONDecodeError, AttributeError):
-        document_id = "Unable to parse response"
-        response_data = response.text
 
+@celery.task(
+    bind=True,
+    autoretry_for=[exceptions.HTTPError],
+    max_retries=settings.celery_max_retries,
+    queue=settings.transcribe_queue,
+)
+def process_audio_transcribe_summarize_v2(
+    self,
+    owner_id: str,
+    filename: str,
+    email: str,
+    sub: str,
+    received_at: float,
+    room: Optional[str],
+    recording_date: Optional[str],
+    recording_time: Optional[str],
+    language: Optional[str],
+    download_link: Optional[str],
+    context_language: Optional[str] = None,
+):
+    """Process an audio file by transcribing it and generating a summary.
+
+    This Celery task orchestrates:
+    1. Audio transcription via WhisperX
+    2. Transcript formatting
+    3. Webhook submission
+    4. Conditional summarization queuing
+
+    Args:
+        self: Celery task instance (passed on with bind=True)
+        owner_id: Unique identifier of the recording owner.
+        filename: Name of the audio file in MinIO storage.
+        email: Email address of the recording owner.
+        sub: OIDC subject identifier of the recording owner.
+        received_at: Unix timestamp when the recording was received.
+        room: room name where the recording took place.
+        recording_date: Date of the recording (localized display string).
+        recording_time: Time of the recording (localized display string).
+        language: ISO 639-1 language code for transcription.
+        download_link: URL to download the original recording.
+        context_language: ISO 639-1 language code of the meeting summary context text.
+    """
     logger.info(
-        "Webhook success | Document %s submitted (HTTP %s)",
-        document_id,
-        response.status_code,
+        "Notification received | Owner: %s | Room: %s",
+        owner_id,
+        room,
     )
-    logger.debug("Full response: %s", response_data)
 
+    task_id = self.request.id
+
+    transcription = transcribe_audio(task_id, filename, language)
+    if transcription is None:
+        return
+
+    content, title = format_transcript(
+        transcription,
+        context_language,
+        language,
+        room,
+        recording_date,
+        recording_time,
+        download_link,
+    )
+
+    submit_content(content, title, email, sub)
     metadata_manager.capture(task_id, settings.posthog_event_success)
 
+    # LLM Summarization
     if (
         analytics.is_feature_enabled("summary-enabled", distinct_id=owner_id)
         and settings.is_summary_enabled
@@ -286,12 +291,11 @@ def summarize_transcription(
     # a singleton client. This is a performance trade-off we accept to ensure per-user
     # privacy controls in observability traces.
     llm_observability = LLMObservability(
-        logger=logger,
         user_has_tracing_consent=user_has_tracing_consent,
         session_id=self.request.id,
         user_id=owner_id,
     )
-    llm_service = LLMService(llm_observability=llm_observability, logger=logger)
+    llm_service = LLMService(llm_observability=llm_observability)
 
     tldr = llm_service.call(PROMPT_SYSTEM_TLDR, transcript, name="tldr")
 
@@ -334,22 +338,9 @@ def summarize_transcription(
     logger.info("Summary cleaned")
 
     summary = tldr + "\n\n" + cleaned_summary + "\n\n" + next_steps
+    summary_title = settings.summary_title_template.format(title=title)
 
-    data = {
-        "title": settings.summary_title_template.format(
-            title=title,
-        ),
-        "content": summary,
-        "email": email,
-        "sub": sub,
-    }
-
-    logger.debug("Submitting webhook to %s", settings.webhook_url)
-
-    response = post_with_retries(settings.webhook_url, data)
-
-    logger.info("Webhook submitted successfully. Status: %s", response.status_code)
-    logger.debug("Response body: %s", response.text)
+    submit_content(summary, summary_title, email, sub)
 
     llm_observability.flush()
     logger.debug("LLM observability flushed")

src/summary/summary/core/config.py

@@ -1,7 +1,7 @@
 """Application configuration and settings."""
 
 from functools import lru_cache
-from typing import Annotated, List, Optional, Set
+from typing import Annotated, List, Literal, Optional, Set
 
 from fastapi import Depends
 from pydantic import SecretStr
@@ -51,7 +51,6 @@ class Settings(BaseSettings):
 
     # Transcription processing
     hallucination_patterns: List[str] = ["Vap'n'Roll Thierry"]
-    hallucination_replacement_text: str = "[Texte impossible à transcrire]"
 
     # Webhook-related settings
     webhook_max_retries: int = 2
@@ -60,11 +59,10 @@ class Settings(BaseSettings):
     webhook_api_token: SecretStr
     webhook_url: str
 
+    # Locale
+    default_context_language: Literal["de", "en", "fr", "nl"] = "fr"
+
     # Output related settings
-    document_default_title: Optional[str] = "Transcription"
-    document_title_template: Optional[str] = (
-        'Réunion "{room}" du {room_recording_date} à {room_recording_time}'
-    )
     summary_title_template: Optional[str] = "Résumé de {title}"
 
     # Summary related settings

src/summary/summary/core/file_service.py

@@ -1,5 +1,6 @@
 """File service to encapsulate files' manipulations."""
 
+import logging
 import os
 import subprocess
 import tempfile
@@ -15,6 +16,9 @@ from summary.core.config import get_settings
 settings = get_settings()
 
 
+logger = logging.getLogger(__name__)
+
+
 class FileServiceException(Exception):
     """Base exception for file service operations."""
 
@@ -24,10 +28,8 @@ class FileServiceException(Exception):
 class FileService:
     """Service for downloading and preparing files from MinIO storage."""
 
-    def __init__(self, logger):
+    def __init__(self):
         """Initialize FileService with MinIO client and configuration."""
-        self._logger = logger
-
         endpoint = (
             settings.aws_s3_endpoint_url.removeprefix("https://")
             .removeprefix("http://")
@@ -53,16 +55,16 @@ class FileService:
         The file is downloaded to a temporary location for local manipulation
         such as validation, conversion, or processing before being used.
         """
-        self._logger.info("Download recording | object_key: %s", remote_object_key)
+        logger.info("Download recording | object_key: %s", remote_object_key)
 
         if not remote_object_key:
-            self._logger.warning("Invalid object_key '%s'", remote_object_key)
+            logger.warning("Invalid object_key '%s'", remote_object_key)
             raise ValueError("Invalid object_key")
 
         extension = Path(remote_object_key).suffix.lower()
 
         if extension not in self._allowed_extensions:
-            self._logger.warning("Invalid file extension '%s'", extension)
+            logger.warning("Invalid file extension '%s'", extension)
             raise ValueError(f"Invalid file extension '{extension}'")
 
         response = None
@@ -81,8 +83,8 @@ class FileService:
                 tmp.flush()
                 local_path = Path(tmp.name)
 
-                self._logger.info("Recording successfully downloaded")
-                self._logger.debug("Recording local file path: %s", local_path)
+                logger.info("Recording successfully downloaded")
+                logger.debug("Recording local file path: %s", local_path)
 
                 return local_path
 
@@ -100,7 +102,7 @@ class FileService:
         file_metadata = mutagen.File(local_path).info
         duration = file_metadata.length
 
-        self._logger.info(
+        logger.info(
             "Recording file duration: %.2f seconds",
             duration,
         )
@@ -109,14 +111,14 @@ class FileService:
             error_msg = "Recording too long. Limit is %.2fs seconds" % (
                 self._max_duration,
             )
-            self._logger.error(error_msg)
+            logger.error(error_msg)
             raise ValueError(error_msg)
 
         return duration
 
     def _extract_audio_from_video(self, video_path: Path) -> Path:
         """Extract audio from video file (e.g., MP4) and save as audio file."""
-        self._logger.info("Extracting audio from video file: %s", video_path)
+        logger.info("Extracting audio from video file: %s", video_path)
 
         with tempfile.NamedTemporaryFile(
             suffix=".m4a", delete=False, prefix="audio_extract_"
@@ -140,16 +142,16 @@ class FileService:
                 command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True
             )
 
-            self._logger.info("Audio successfully extracted to: %s", output_path)
+            logger.info("Audio successfully extracted to: %s", output_path)
             return output_path
 
         except FileNotFoundError as e:
-            self._logger.error("ffmpeg not found. Please install ffmpeg.")
+            logger.error("ffmpeg not found. Please install ffmpeg.")
             if output_path.exists():
                 os.remove(output_path)
             raise RuntimeError("ffmpeg is not installed or not in PATH") from e
         except subprocess.CalledProcessError as e:
-            self._logger.error("Audio extraction failed: %s", e.stderr.decode())
+            logger.error("Audio extraction failed: %s", e.stderr.decode())
             if output_path.exists():
                 os.remove(output_path)
             raise RuntimeError("Failed to extract audio.") from e
@@ -173,7 +175,7 @@ class FileService:
             extension = downloaded_path.suffix.lower()
 
             if extension in settings.recording_video_extensions:
-                self._logger.info("Video file detected, extracting audio...")
+                logger.info("Video file detected, extracting audio...")
                 extracted_audio_path = self._extract_audio_from_video(downloaded_path)
                 processed_path = extracted_audio_path
             else:
@@ -194,8 +196,6 @@ class FileService:
 
                 try:
                     os.remove(path)
-                    self._logger.debug("Temporary file removed: %s", path)
+                    logger.debug("Temporary file removed: %s", path)
                 except OSError as e:
-                    self._logger.warning(
-                        "Failed to remove temporary file %s: %s", path, e
-                    )
+                    logger.warning("Failed to remove temporary file %s: %s", path, e)

src/summary/summary/core/llm_service.py

@@ -1,5 +1,6 @@
 """LLM service to encapsulate LLM's calls."""
 
+import logging
 from typing import Any, Mapping, Optional
 
 import openai
@@ -10,6 +11,9 @@ from summary.core.config import get_settings
 settings = get_settings()
 
 
+logger = logging.getLogger(__name__)
+
+
 class LLMObservability:
     """Manage observability and tracing for LLM calls using Langfuse.
 
@@ -21,13 +25,11 @@ class LLMObservability:
 
     def __init__(
         self,
-        logger,
         session_id: str,
         user_id: str,
         user_has_tracing_consent: bool = False,
     ):
         """Initialize the LLMObservability client."""
-        self._logger = logger
         self._observability_client: Optional[Langfuse] = None
         self.session_id = session_id
         self.user_id = user_id
@@ -75,7 +77,7 @@ class LLMObservability:
         }
 
         if not self.is_enabled:
-            self._logger.debug("Using regular OpenAI client (observability disabled)")
+            logger.debug("Using regular OpenAI client (observability disabled)")
             return openai.OpenAI(**base_args)
 
         # Langfuse's OpenAI wrapper is imported here to avoid triggering client
@@ -83,7 +85,7 @@ class LLMObservability:
         # is missing. Conditional import ensures Langfuse only initializes when enabled.
         from langfuse.openai import openai as langfuse_openai  # noqa: PLC0415
 
-        self._logger.debug("Using LangfuseOpenAI client (observability enabled)")
+        logger.debug("Using LangfuseOpenAI client (observability enabled)")
         return langfuse_openai.OpenAI(**base_args)
 
     def flush(self):
@@ -99,11 +101,10 @@ class LLMException(Exception):
 class LLMService:
     """Service for performing calls to the LLM configured in the settings."""
 
-    def __init__(self, llm_observability, logger):
+    def __init__(self, llm_observability):
         """Init the LLMService once."""
         self._client = llm_observability.get_openai_client()
         self._observability = llm_observability
-        self._logger = logger
 
     def call(
         self,
@@ -140,5 +141,5 @@ class LLMService:
             return response.choices[0].message.content
 
         except Exception as e:
-            self._logger.exception("LLM call failed: %s", e)
+            logger.exception("LLM call failed: %s", e)
             raise LLMException(f"LLM call failed: {e}") from e

src/summary/summary/core/locales/__init__.py

@@ -0,0 +1,30 @@
+"""Locale support for the summary service."""
+
+from typing import Optional
+
+from summary.core.config import get_settings
+from summary.core.locales import de, en, fr, nl
+from summary.core.locales.strings import LocaleStrings
+
+_LOCALES = {"fr": fr, "en": en, "de": de, "nl": nl}
+
+
+def get_locale(*languages: Optional[str]) -> LocaleStrings:
+    """Return locale strings for the first matching language candidate.
+
+    Accept language codes in decreasing priority order and return the
+    locale for the first one that matches a known locale.
+    Fall back to the configured default_context_language.
+    """
+    for lang in languages:
+        if not lang:
+            continue
+        if lang in _LOCALES:
+            return _LOCALES[lang].STRINGS
+
+        # Provide fallback for longer formats of ISO 639-1 (e.g. "en-au" -> "en")
+        base_lang = lang.split("-")[0]
+        if base_lang in _LOCALES:
+            return _LOCALES[base_lang].STRINGS
+
+    return _LOCALES[get_settings().default_context_language].STRINGS

src/summary/summary/core/locales/de.py

@@ -0,0 +1,34 @@
+"""German locale strings."""
+
+from summary.core.locales.strings import LocaleStrings
+
+STRINGS = LocaleStrings(
+    empty_transcription="""
+**In Ihrer Transkription wurde kein Audioinhalt erkannt.**
+
+*Wenn Sie glauben, dass es sich um einen Fehler handelt, zögern Sie nicht,
+unseren technischen Support zu kontaktieren: visio@numerique.gouv.fr*
+
+.
+
+.
+
+.
+
+Einige Punkte, die wir Ihnen empfehlen zu überprüfen:
+- War ein Mikrofon aktiviert?
+- Waren Sie nah genug am Mikrofon?
+- Ist das Mikrofon von guter Qualität?
+- Dauert die Aufnahme länger als 30 Sekunden?
+
+""",
+    download_header_template=(
+        "\n*Laden Sie Ihre Aufnahme herunter, "
+        "indem Sie [diesem Link folgen]({download_link})*\n"
+    ),
+    hallucination_replacement_text="[Text konnte nicht transkribiert werden]",
+    document_default_title="Transkription",
+    document_title_template=(
+        'Besprechung "{room}" am {room_recording_date} um {room_recording_time}'
+    ),
+)

src/summary/summary/core/locales/en.py

@@ -0,0 +1,33 @@
+"""English locale strings."""
+
+from summary.core.locales.strings import LocaleStrings
+
+STRINGS = LocaleStrings(
+    empty_transcription="""
+**No audio content was detected in your transcription.**
+
+*If you believe this is an error, please do not hesitate to contact
+our technical support: visio@numerique.gouv.fr*
+
+.
+
+.
+
+.
+
+A few things we recommend you check:
+- Was a microphone enabled?
+- Were you close enough to the microphone?
+- Is the microphone of good quality?
+- Is the recording longer than 30 seconds?
+
+""",
+    download_header_template=(
+        "\n*Download your recording by [following this link]({download_link})*\n"
+    ),
+    hallucination_replacement_text="[Unable to transcribe text]",
+    document_default_title="Transcription",
+    document_title_template=(
+        'Meeting "{room}" on {room_recording_date} at {room_recording_time}'
+    ),
+)

src/summary/summary/core/locales/fr.py

@@ -0,0 +1,33 @@
+"""French locale strings (default)."""
+
+from summary.core.locales.strings import LocaleStrings
+
+STRINGS = LocaleStrings(
+    empty_transcription="""
+**Aucun contenu audio n'a été détecté dans votre transcription.**
+
+*Si vous pensez qu'il s'agit d'une erreur, n'hésitez pas à contacter
+notre support technique : visio@numerique.gouv.fr*
+
+.
+
+.
+
+.
+
+Quelques points que nous vous conseillons de vérifier :
+- Un micro était-il activé ?
+- Étiez-vous suffisamment proche ?
+- Le micro est-il de bonne qualité ?
+- L'enregistrement dure-t-il plus de 30 secondes ?
+
+""",
+    download_header_template=(
+        "\n*Télécharger votre enregistrement en [suivant ce lien]({download_link})*\n"
+    ),
+    hallucination_replacement_text="[Texte impossible à transcrire]",
+    document_default_title="Transcription",
+    document_title_template=(
+        'Réunion "{room}" du {room_recording_date} à {room_recording_time}'
+    ),
+)

src/summary/summary/core/locales/nl.py

@@ -0,0 +1,33 @@
+"""Dutch locale strings."""
+
+from summary.core.locales.strings import LocaleStrings
+
+STRINGS = LocaleStrings(
+    empty_transcription="""
+**Er is geen audio-inhoud gedetecteerd in uw transcriptie.**
+
+*Als u denkt dat dit een fout is, aarzel dan niet om contact op te nemen
+met onze technische ondersteuning: visio@numerique.gouv.fr*
+
+.
+
+.
+
+.
+
+Een paar punten die wij u aanraden te controleren:
+- Was er een microfoon ingeschakeld?
+- Was u dicht genoeg bij de microfoon?
+- Is de microfoon van goede kwaliteit?
+- Duurt de opname langer dan 30 seconden?
+
+""",
+    download_header_template=(
+        "\n*Download uw opname door [deze link te volgen]({download_link})*\n"
+    ),
+    hallucination_replacement_text="[Tekst kon niet worden getranscribeerd]",
+    document_default_title="Transcriptie",
+    document_title_template=(
+        'Vergadering "{room}" op {room_recording_date} om {room_recording_time}'
+    ),
+)

src/summary/summary/core/locales/strings.py

@@ -0,0 +1,15 @@
+"""Locale types for the summary service."""
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class LocaleStrings:
+    """All translatable output strings for the summary pipeline."""
+
+    # transcript_formatter.py
+    empty_transcription: str
+    download_header_template: str
+    hallucination_replacement_text: str
+    document_default_title: str
+    document_title_template: str

src/summary/summary/core/transcript_formatter.py

@@ -4,34 +4,13 @@ import logging
 from typing import Optional, Tuple
 
 from summary.core.config import get_settings
+from summary.core.locales import LocaleStrings
 
 settings = get_settings()
 
 logger = logging.getLogger(__name__)
 
 
-DEFAULT_EMPTY_TRANSCRIPTION = """
-**Aucun contenu audio n’a été détecté dans votre transcription.**
-
-
-*Si vous pensez qu’il s’agit d’une erreur, n’hésitez pas à contacter
-notre support technique : visio@numerique.gouv.fr*
-
-.
-
-.
-
-.
-
-Quelques points que nous vous conseillons de vérifier :
-- Un micro était-il activé ?
-- Étiez-vous suffisamment proche ?
-- Le micro est-il de bonne qualité ?
-- L’enregistrement dure-t-il plus de 30 secondes ?
-
-"""
-
-
 class TranscriptFormatter:
     """Formats WhisperX transcription output into readable conversation format.
 
@@ -42,12 +21,10 @@ class TranscriptFormatter:
     - Generating descriptive titles from context
     """
 
-    def __init__(self):
-        """Initialize formatter with settings."""
+    def __init__(self, locale: LocaleStrings):
+        """Initialize formatter with settings and locale."""
         self.hallucination_patterns = settings.hallucination_patterns
-        self.hallucination_replacement_text = settings.hallucination_replacement_text
-        self.default_title = settings.document_default_title
-        self.default_empty_transcription = DEFAULT_EMPTY_TRANSCRIPTION
+        self._locale = locale
 
     def _get_segments(self, transcription):
         """Extract segments from transcription object or dictionary."""
@@ -71,7 +48,7 @@ class TranscriptFormatter:
         segments = self._get_segments(transcription)
 
         if not segments:
-            content = self.default_empty_transcription
+            content = self._locale.empty_transcription
         else:
             content = self._format_speaker(segments)
             content = self._remove_hallucinations(content)
@@ -83,7 +60,7 @@ class TranscriptFormatter:
 
     def _remove_hallucinations(self, content: str) -> str:
         """Remove hallucination patterns from content."""
-        replacement = self.hallucination_replacement_text or ""
+        replacement = self._locale.hallucination_replacement_text or ""
 
         for pattern in self.hallucination_patterns:
             content = content.replace(pattern, replacement)
@@ -111,9 +88,8 @@ class TranscriptFormatter:
         if not download_link:
             return content
 
-        header = (
-            f"\n*Télécharger votre enregistrement "
-            f"en [suivant ce lien]({download_link})*\n"
+        header = self._locale.download_header_template.format(
+            download_link=download_link
         )
         content = header + content
 
@@ -127,9 +103,9 @@ class TranscriptFormatter:
     ) -> str:
         """Generate title from context or return default."""
         if not room or not recording_date or not recording_time:
-            return self.default_title
+            return self._locale.document_default_title
 
-        return settings.document_title_template.format(
+        return self._locale.document_title_template.format(
             room=room,
             room_recording_date=recording_date,
             room_recording_time=recording_time,

src/summary/summary/core/webhook_service.py

@@ -0,0 +1,73 @@
+"""Service for delivering content to external destinations."""
+
+import json
+import logging
+
+from requests import Session
+from requests.adapters import HTTPAdapter
+from urllib3.util import Retry
+
+from summary.core.config import get_settings
+
+settings = get_settings()
+
+logger = logging.getLogger(__name__)
+
+
+def _create_retry_session():
+    """Create an HTTP session configured with retry logic."""
+    session = Session()
+    retries = Retry(
+        total=settings.webhook_max_retries,
+        backoff_factor=settings.webhook_backoff_factor,
+        status_forcelist=settings.webhook_status_forcelist,
+        allowed_methods={"POST"},
+    )
+    session.mount("https://", HTTPAdapter(max_retries=retries))
+    return session
+
+
+def _post_with_retries(url, data):
+    """Send POST request with automatic retries."""
+    session = _create_retry_session()
+    session.headers.update(
+        {"Authorization": f"Bearer {settings.webhook_api_token.get_secret_value()}"}
+    )
+    try:
+        response = session.post(url, json=data)
+        response.raise_for_status()
+        return response
+    finally:
+        session.close()
+
+
+def submit_content(content, title, email, sub):
+    """Submit content to the configured webhook destination.
+
+    Builds the payload, sends it with retries, and logs the outcome.
+    """
+    data = {
+        "title": title,
+        "content": content,
+        "email": email,
+        "sub": sub,
+    }
+
+    logger.debug("Submitting to %s", settings.webhook_url)
+    logger.debug("Request payload: %s", json.dumps(data, indent=2))
+
+    response = _post_with_retries(settings.webhook_url, data)
+
+    try:
+        response_data = response.json()
+        document_id = response_data.get("id", "N/A")
+    except (json.JSONDecodeError, AttributeError):
+        document_id = "Unable to parse response"
+        response_data = response.text
+
+    logger.info(
+        "Delivery success | Document %s submitted (HTTP %s)",
+        document_id,
+        response.status_code,
+    )
+    logger.debug("Full response: %s", response_data)