Summarize tone: re-summarize in a chosen voice (backend)
POST /reprocess?job=summarize&tone=<voice> persists the voice on the job row (queue carries only ids, so a sweep re-enqueue keeps it), the LLM prompt appends 'write every field in a <tone> tone — the tone colors the wording, never the facts', and the resulting summary records its tone (SummaryOut.tone). Plain re-summarize clears a previous tone. Migration tone0000000001 (summaries.tone, processing_jobs.tone). 78 passed, 1 skipped; ruff clean.
This commit is contained in:
parent
0d8f3be614
commit
b63f49241d
11 changed files with 82 additions and 13 deletions
|
|
@ -112,7 +112,8 @@ class SummaryResult:
|
|||
class LlmProvider(Protocol):
|
||||
name: str
|
||||
|
||||
async def summarize(self, transcript: str, *, title: str | None = None) -> SummaryResult: ...
|
||||
async def summarize(self, transcript: str, *, title: str | None = None,
|
||||
tone: str | None = None) -> SummaryResult: ...
|
||||
|
||||
|
||||
def get_transcription_provider(settings: Settings) -> TranscriptionProvider | None:
|
||||
|
|
|
|||
|
|
@ -21,12 +21,21 @@ SYSTEM_PROMPT = (
|
|||
MAX_TRANSCRIPT_CHARS = 12_000
|
||||
|
||||
|
||||
def build_user_message(transcript: str, title: str | None) -> str:
|
||||
def build_user_message(transcript: str, title: str | None,
|
||||
tone: str | None = None) -> str:
|
||||
text = transcript[:MAX_TRANSCRIPT_CHARS]
|
||||
if len(transcript) > MAX_TRANSCRIPT_CHARS:
|
||||
text += f"\n\n[truncated from {len(transcript)} chars]"
|
||||
head = f'Title: "{title}"\n\n' if title else ""
|
||||
return head + "Transcript:\n" + text
|
||||
msg = head + "Transcript:\n" + text
|
||||
if tone:
|
||||
# Same JSON contract and same fidelity rules — only the voice changes.
|
||||
msg += (
|
||||
f"\n\nWrite every field of the JSON in a {tone} tone of voice. "
|
||||
"Stay faithful to the transcript: the tone colors the wording, "
|
||||
"never the facts."
|
||||
)
|
||||
return msg
|
||||
|
||||
|
||||
def parse_summary(data: object, model: str) -> SummaryResult:
|
||||
|
|
|
|||
|
|
@ -38,7 +38,8 @@ class OllamaProvider:
|
|||
async with httpx.AsyncClient(timeout=self.timeout_s) as client:
|
||||
yield client
|
||||
|
||||
async def summarize(self, transcript: str, *, title: str | None = None) -> SummaryResult:
|
||||
async def summarize(self, transcript: str, *, title: str | None = None,
|
||||
tone: str | None = None) -> SummaryResult:
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"stream": False,
|
||||
|
|
@ -51,7 +52,8 @@ class OllamaProvider:
|
|||
"options": {"num_ctx": 8192},
|
||||
"messages": [
|
||||
{"role": "system", "content": SYSTEM_PROMPT},
|
||||
{"role": "user", "content": build_user_message(transcript, title)},
|
||||
{"role": "user",
|
||||
"content": build_user_message(transcript, title, tone)},
|
||||
],
|
||||
}
|
||||
try:
|
||||
|
|
|
|||
|
|
@ -42,7 +42,8 @@ class OpenAICompatProvider:
|
|||
async with httpx.AsyncClient(timeout=self.timeout_s) as client:
|
||||
yield client
|
||||
|
||||
async def summarize(self, transcript: str, *, title: str | None = None) -> SummaryResult:
|
||||
async def summarize(self, transcript: str, *, title: str | None = None,
|
||||
tone: str | None = None) -> SummaryResult:
|
||||
headers = (
|
||||
{"Authorization": f"Bearer {self.api_key}"} if self.api_key else {}
|
||||
)
|
||||
|
|
@ -52,7 +53,8 @@ class OpenAICompatProvider:
|
|||
"response_format": {"type": "json_object"},
|
||||
"messages": [
|
||||
{"role": "system", "content": SYSTEM_PROMPT},
|
||||
{"role": "user", "content": build_user_message(transcript, title)},
|
||||
{"role": "user",
|
||||
"content": build_user_message(transcript, title, tone)},
|
||||
],
|
||||
}
|
||||
try:
|
||||
|
|
|
|||
|
|
@ -437,12 +437,14 @@ async def run_summarize(ctx: dict, recording_id: str) -> None:
|
|||
rec.processing_status = ProcessingStatus.processing
|
||||
await session.commit() # visible before the long LLM call
|
||||
try:
|
||||
result = await provider.summarize(text, title=rec.title)
|
||||
result = await provider.summarize(text, title=rec.title,
|
||||
tone=job.tone)
|
||||
except AIError as e:
|
||||
await _fail(session, rec, job, str(e), ctx, e)
|
||||
await session.commit()
|
||||
return
|
||||
await store_summary(session, rec, result.to_dict(), provider.name, result.model)
|
||||
await store_summary(session, rec, result.to_dict(), provider.name,
|
||||
result.model, tone=job.tone)
|
||||
job.status = JobStatus.succeeded
|
||||
job.stage = None
|
||||
job.progress = 100
|
||||
|
|
@ -506,6 +508,7 @@ async def store_summary(
|
|||
content: dict,
|
||||
provider_name: str,
|
||||
model: str,
|
||||
tone: str | None = None,
|
||||
) -> None:
|
||||
existing = (
|
||||
await session.scalars(
|
||||
|
|
@ -531,6 +534,7 @@ async def store_summary(
|
|||
version=(max_version or 0) + 1,
|
||||
provider=provider_name,
|
||||
model=model,
|
||||
tone=tone,
|
||||
content=content,
|
||||
edited_by_user=False,
|
||||
)
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue