"""FastAPI service for Verifiable Dictation Suite.

Full pipeline: audio → STT → claim extraction → QKL lattice → XGML graph
→ Pi Sheriff Node attestation → XAL envelope.

Aevion LLC | SDVOSB | CAGE 15NV7
"""

import hashlib
import uuid
from pathlib import Path
import sys

from fastapi import FastAPI, UploadFile
from fastapi.responses import JSONResponse

# Support both package and standalone imports for HF Spaces / Vercel
try:
    from avds.attestation import generate_receipt, verify_receipt
    from avds.translators.qkl import build_qkl_lattice
    from avds.translators.xal import wrap_as_xal
    from avds.translators.xgml import build_xgml_graph
except ImportError:
    sys.path.insert(0, str(Path(__file__).parent))
    from attestation import generate_receipt, verify_receipt
    from translators.qkl import build_qkl_lattice
    from translators.xal import wrap_as_xal
    from translators.xgml import build_xgml_graph

app = FastAPI(title="AVDS — Verifiable Dictation Suite")

# Session storage (in-memory for single-instance; use Redis for production)
sessions: dict[str, dict] = {}


@app.post("/transcribe")
async def transcribe(file: UploadFile) -> JSONResponse:
    """Accept audio, run full AVDS pipeline, return XAL envelope."""
    content = await file.read()
    audio_hash = hashlib.sha256(content).hexdigest()
    session_id = f"avds-{uuid.uuid4().hex[:12]}"

    # Step 1: Transcribe (currently mock; swap for Voxtral/Mistral STT)
    # transcript = await call_stt_api(content, file.content_type)
    transcript = f"[MOCK TRANSCRIPT for {audio_hash[:8]}]"

    # Step 2: Generate hardware-attested receipt
    stt_model = "voxtral-mini-2507"
    receipt = await generate_receipt(audio_hash, transcript, stt_model)

    # Step 3: Extract claims from transcript
    claims = _extract_claims(transcript)

    # Step 4: Build QKL lattice
    qkl = build_qkl_lattice(claims, session_id)

    # Step 5: Build XGML audit graph
    xgml = build_xgml_graph(
        session_id=session_id,
        audio_hash=audio_hash,
        transcript=transcript,
        stt_model=stt_model,
        receipt=receipt,
        claims_count=len(claims),
    )

    # Step 6: Wrap in XAL envelope
    xal = wrap_as_xal(transcript, receipt)
    xal["qkl"] = qkl
    xal["xgml"] = xgml

    # Store session
    sessions[session_id] = {
        "audio_hash": audio_hash,
        "transcript": transcript,
        "receipt": receipt,
        "claims": claims,
        "qkl": qkl,
        "xgml": xgml,
    }

    return JSONResponse(content=xal)


@app.post("/verify")
async def verify(receipt: dict) -> dict:
    """Verify a receipt's signature and integrity."""
    valid = verify_receipt(receipt)
    return {"valid": valid, "receipt_id": receipt.get("receipt_id")}


@app.get("/health")
async def health() -> dict:
    """Health check endpoint."""
    return {
        "status": "ok",
        "sessions": len(sessions),
        "pipeline": ["stt", "receipt", "claims", "qkl", "xgml", "xal"],
    }


@app.get("/session/{session_id}")
async def get_session(session_id: str) -> JSONResponse:
    """Retrieve a stored session by ID."""
    if session_id in sessions:
        return JSONResponse(content=sessions[session_id])
    return JSONResponse(content={"error": "session not found"}, status_code=404)


def _extract_claims(transcript: str) -> list[dict]:
    """Extract structured claims from a transcript.

    Currently returns mock claims. In production, this calls a
    cleanup LLM (Qwen2.5 or similar) to extract factual claims
    with confidence scores.
    """
    if not transcript or transcript.startswith("[MOCK"):
        return [{"text": "mock claim", "confidence": 0.5, "source": "stt"}]
    # Split on sentence boundaries for claim extraction
    sentences = [s.strip() for s in transcript.split(".") if s.strip()]
    return [
        {"text": s, "confidence": 0.7, "source": "stt"}
        for s in sentences[:10]  # Cap at 10 claims
    ]