src/server/services/summarization_service.py

from __future__ import annotations

from typing import Dict, Iterable

from src.summarization import summarize_transcript, generate_title

from ..models.summarization import SummaryRequest


def iter_summary_events(payload: SummaryRequest) -> Iterable[Dict[str, str]]:
    try:
        # Generate title if requested
        title = None
        if payload.generate_title:
            title = generate_title(payload.transcript, payload.llm_model)
            yield {"type": "title", "content": title}

        # Generate summary
        generator = summarize_transcript(
            transcript=payload.transcript,
            selected_gguf_model=payload.llm_model,
            prompt_input=payload.prompt,
        )

        for chunk in generator:
            yield {"type": "partial", "content": chunk}

        yield {"type": "complete"}

    except Exception as exc:
        # Response already started (HTTP 200 + NDJSON); surface failures as a
        # terminal error event instead of a raise that truncates the stream.
        yield {"type": "error", "message": f"Summary failed: {exc}"}