app/llm/generation_facade.py

"""Unified async entry points for section generation LLM calls.

Callers in async handlers should use these helpers instead of branching on
``enable_async_pipeline`` at every call site.
"""

from __future__ import annotations

from typing import TYPE_CHECKING

from app.async_executor import run_sync_in_executor
from app.config import settings
from app.generator.adapter import get_llm_adapter
from app.llm.async_llm_adapter import get_async_llm_adapter

if TYPE_CHECKING:
    from app.models.schemas import WritingStyleProfile


async def generate_section(
    *,
    skeleton: str,
    bullets: list[str],
    snippets: list[str],
    style_profile: "WritingStyleProfile | None" = None,
    temperature: float = 0.2,
    creativity_hint: str = "",
    document_context: list[str] | None = None,
    style_anchor: str | None = None,
    hierarchy_section_snippets: list[str] | None = None,
    paragraph_snippets: list[str] | None = None,
    identity_facts: str | None = None,
    survey_level: int | None = None,
    reference_only_context: bool = False,
    ai_percent: int | None = None,
    interference_level: str | None = None,
    tenant_id: str | None = None,
    scope_fence: str | None = None,
) -> str:
    if settings.enable_async_pipeline:
        return await get_async_llm_adapter().generate_section(
            skeleton=skeleton,
            bullets=bullets,
            snippets=snippets,
            style_profile=style_profile,
            temperature=temperature,
            creativity_hint=creativity_hint,
            document_context=document_context,
            style_anchor=style_anchor,
            hierarchy_section_snippets=hierarchy_section_snippets,
            paragraph_snippets=paragraph_snippets,
            identity_facts=identity_facts,
            survey_level=survey_level,
            reference_only_context=reference_only_context,
            ai_percent=ai_percent,
            interference_level=interference_level,
            tenant_id=tenant_id,
            scope_fence=scope_fence,
        )
    return await run_sync_in_executor(
        get_llm_adapter().generate_section,
        skeleton=skeleton,
        bullets=bullets,
        snippets=snippets,
        style_profile=style_profile,
        temperature=temperature,
        creativity_hint=creativity_hint,
        document_context=document_context,
        style_anchor=style_anchor,
        hierarchy_section_snippets=hierarchy_section_snippets,
        paragraph_snippets=paragraph_snippets,
        identity_facts=identity_facts,
        survey_level=survey_level,
        reference_only_context=reference_only_context,
        ai_percent=ai_percent,
        interference_level=interference_level,
        tenant_id=tenant_id,
        scope_fence=scope_fence,
    )


async def proofread(
    *,
    text: str,
    bullets: list[str],
    style_profile: "WritingStyleProfile | None" = None,
    temperature: float = 0.15,
    creativity_hint: str = "",
) -> str:
    if settings.enable_async_pipeline:
        return await get_async_llm_adapter().proofread(
            text=text,
            bullets=bullets,
            style_profile=style_profile,
            temperature=temperature,
            creativity_hint=creativity_hint,
        )
    return await run_sync_in_executor(
        get_llm_adapter().proofread,
        text=text,
        bullets=bullets,
        style_profile=style_profile,
        temperature=temperature,
        creativity_hint=creativity_hint,
    )


async def enhance(
    *,
    text: str,
    bullets: list[str],
    snippets: list[str],
    style_profile: "WritingStyleProfile | None" = None,
    temperature: float = 0.2,
    creativity_hint: str = "",
) -> str:
    if settings.enable_async_pipeline:
        return await get_async_llm_adapter().enhance(
            text=text,
            bullets=bullets,
            snippets=snippets,
            style_profile=style_profile,
            temperature=temperature,
            creativity_hint=creativity_hint,
        )
    return await run_sync_in_executor(
        get_llm_adapter().enhance,
        text=text,
        bullets=bullets,
        snippets=snippets,
        style_profile=style_profile,
        temperature=temperature,
        creativity_hint=creativity_hint,
    )


async def validate_section_compliance(
    *,
    survey_level: int | None,
    section_code: str,
    bullets: list[str],
    evidence_snippets: list[str],
    text: str,
) -> str:
    if settings.enable_async_pipeline:
        return await get_async_llm_adapter().validate_section_compliance(
            survey_level=survey_level,
            section_code=section_code,
            bullets=bullets,
            evidence_snippets=evidence_snippets,
            text=text,
        )
    return await run_sync_in_executor(
        get_llm_adapter().validate_section_compliance,
        survey_level=survey_level,
        section_code=section_code,
        bullets=bullets,
        evidence_snippets=evidence_snippets,
        text=text,
    )


async def constrained_weave(
    *,
    section_code: str,
    section_title: str | None,
    bullets: list[str],
    standard_passages: list[str],
    survey_level: int | None = None,
    tenant_id: str | None = None,
) -> str:
    if settings.enable_async_pipeline:
        return await get_async_llm_adapter().constrained_weave(
            section_code=section_code,
            section_title=section_title,
            bullets=bullets,
            standard_passages=standard_passages,
            survey_level=survey_level,
            tenant_id=tenant_id,
        )
    return await run_sync_in_executor(
        get_llm_adapter().constrained_weave,
        section_code=section_code,
        section_title=section_title,
        bullets=bullets,
        standard_passages=standard_passages,
        survey_level=survey_level,
        tenant_id=tenant_id,
    )