src/pipeline/models.py

# YJ/pipeline/models.py

from __future__ import annotations
from typing import List, Optional, Literal, Dict, Any
from uuid import uuid4

from pydantic import BaseModel, Field


class LanguageAssessment(BaseModel):
    tool: Optional[str] = None  # REVT, SELSI 등
    result_level: Optional[str] = None  # e.g. "mild_delay", "within_range"
    notes: Optional[str] = None


class DevelopmentHistory(BaseModel):
    preterm: Optional[bool] = None
    diagnosed_conditions: List[str] = Field(default_factory=list)
    hearing_issues: Optional[bool] = None


class LanguageEnvironment(BaseModel):
    languages_exposed: List[str] = Field(default_factory=lambda: ["Korean"])
    caregivers: Optional[int] = None
    caregiver_speech_style: Optional[str] = None  # e.g. "many_directives_few_questions"


class BaselineCommunication(BaseModel):
    echolalia_frequency: Optional[str] = None  # "rare", "sometimes", "often"
    spontaneous_utterance_level: Optional[str] = None  # "single_words", "two_words", ...
    scripted_phrases_present: Optional[bool] = None


class UserProfile(BaseModel):
    user_id: str
    child_nickname: Optional[str] = None
    age_months: Optional[int] = None
    sex: Optional[Literal["male", "female", "unspecified"]] = "unspecified"

    language_assessment: Optional[LanguageAssessment] = None
    development_history: Optional[DevelopmentHistory] = None
    language_environment: Optional[LanguageEnvironment] = None
    baseline_communication: Optional[BaselineCommunication] = None


class DialogueTurn(BaseModel):
    speaker: Literal["caregiver", "child", "other"]
    utterance: str
    turn_index: int


class UtteranceAnalysisRequest(BaseModel):
    user_id: str
    context_description: str
    dialogue_log: List[Dict[str, str]]  # raw 입력 (speaker, utterance)
    timestamp: Optional[str] = None


# --- 중간 단계 출력 모델들 ---

class EcholaliaDetectionResult(BaseModel):
    is_echolalia: bool
    type: Optional[Literal["immediate", "delayed", "mixed", "uncertain"]] = None
    completeness: Optional[Literal["full", "partial", "transformed"]] = None
    functional: Optional[bool] = None
    evidence: Optional[str] = None


class EcholaliaFunction(BaseModel):
    tag: Literal[
        "FILLING_TURN",
        "SAYING_YES",
        "LABELING_PRACTICE",
        "SELF_SOOTHING_ROUTINE",
        "STIMMING_VERBAL",
        "SOUND_PLEASURE",
        "LANGUAGE_PROCESSING_OUTLOUD",
        "SITUATION_SCRIPT_REPLAY",
        "COMMUNICATION_ATTEMPT",
        "ANXIETY_DEFENSE",
    ]
    confidence: float
    caregiver_explanation: str


class MeaningInferenceResult(BaseModel):
    echolalia_functions: List[EcholaliaFunction]
    primary_interpretation: str


class DomainProfile(BaseModel):
    summary: str
    relative_level: Literal[
        "strength",
        "age_appropriate",
        "age_appropriate_or_mild_delay",
        "developing",
        "emerging",
        "needs_more_observation",
    ]
    notes: Optional[str] = None


class DevelopmentProfile(BaseModel):
    age_reference: Optional[int] = None
    domains: Dict[str, DomainProfile]
    safety_note: str


class ParentGuide(BaseModel):
    summary_line: str
    immediate_response: List[str]
    modeling_utterance: List[str]
    long_term_tips: List[str]


class SafetyFlags(BaseModel):
    contains_diagnostic_language: bool
    needs_professional_referral_suggestion: bool
    sanitized_text: Optional[Dict[str, Any]] = None  # 수정된 parent_guide 등


class UtteranceAnalysisResult(BaseModel):
    meta: Dict[str, Any]
    echolalia_detection: EcholaliaDetectionResult
    meaning_inference: MeaningInferenceResult
    development_profile: DevelopmentProfile
    parent_guide: ParentGuide
    safety_flags: SafetyFlags

    @staticmethod
    def new_meta(user_id: str, model_version: str = "coconut-v0.1") -> Dict[str, Any]:
        return {
            "user_id": user_id,
            "analysis_id": str(uuid4()),
            "model_version": model_version,
        }