backend/app/api/analysis.py

import uuid
import os
from fastapi import APIRouter, Depends, HTTPException
from sqlalchemy.ext.asyncio import AsyncSession
from sqlalchemy import select

from app.database import get_db
from app.auth import get_current_user
from app.models.user import User
from app.models.chat import ChatMessage
from app.models.evaluation import EvaluationResult
from app.models.metrics import Metrics
from app.services.analysis import RetrievalAnalyzer

router = APIRouter(prefix="/api/analysis", tags=["analysis"])

@router.get("/{message_id}")
async def analyze_message(message_id: uuid.UUID, current_user: User = Depends(get_current_user), db: AsyncSession = Depends(get_db)):
    # 1. Fetch the message natively mapped
    stmt_msg = select(ChatMessage).where(ChatMessage.id == message_id, ChatMessage.user_id == current_user.id)
    msg = (await db.execute(stmt_msg)).scalars().first()
    if not msg:
        raise HTTPException(status_code=404, detail="Message not found")
        
    if msg.role != "assistant":
        raise HTTPException(status_code=400, detail="Analysis is logically only available for assistant messages")
        
    # 2. Fetch associated metrics (if any timer payloads caught mapped traces securely)
    stmt = select(Metrics).where(Metrics.message_id == message_id)
    metrics_record = (await db.execute(stmt)).scalars().first()

    eval_stmt = (
        select(EvaluationResult)
        .where(EvaluationResult.message_id == message_id)
        .order_by(EvaluationResult.created_at.desc())
    )
    eval_record = (await db.execute(eval_stmt)).scalars().first()
    
    # 3. Parse retrieved chunks and instantly leverage our abstract Analyzer class seamlessly
    chunks_data = msg.retrieved_chunks or []
    confidence_threshold = float(os.getenv("ANALYSIS_CONFIDENCE_THRESHOLD", "0.5"))
    analyzer = RetrievalAnalyzer(chunks_data, confidence_threshold=confidence_threshold)
    summary = analyzer.summary_stats()
    
    # Construct structured insight payload natively mapped
    return {
        "message_id": message_id,
        "total_chunks_retrieved": summary["total_chunks_retrieved"],
        "avg_similarity": summary["avg_similarity"],
        "warning_flag": summary["warning_flag"],
        "confidence_threshold": summary["confidence_threshold"],
        "score_distribution": summary["score_distribution"],
        "chunk_diversity": summary["chunk_diversity"],
        "top_contributors": summary["top_contributors"],
        "ranked_chunks": summary["ranked_chunks"],
        "summary_stats": summary,
        "evaluation": {
            "faithfulness": eval_record.faithfulness if eval_record else None,
            "answer_relevancy": eval_record.answer_relevancy if eval_record else None,
            "context_precision": eval_record.context_precision if eval_record else None,
            "context_recall": eval_record.context_recall if eval_record else None,
        } if eval_record else None,
        "timing_breakdown": {
            "chunking_time_ms": metrics_record.chunking_time_ms if metrics_record else None,
            "embedding_time_ms": metrics_record.embedding_time_ms if metrics_record else None,
            "retrieval_time_ms": metrics_record.retrieval_time_ms if metrics_record else None,
            "reranking_time_ms": metrics_record.reranking_time_ms if metrics_record else None,
            "llm_time_ms": metrics_record.llm_time_ms if metrics_record else None,
            "total_time_ms": metrics_record.total_time_ms if metrics_record else None,
        } if metrics_record else None
    }