src/profiling/python-service/main.py

"""
FastAPI Profiling Service v2 — NVIDIA NIM powered.

Endpoints:
  POST /profile  → Profile company + compute score (single pipeline)
  GET  /health   → Service health check

Security:
  Bearer token authentication (shared secret with Node.js orchestration layer)
"""

import logging
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException, Depends
from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
from pydantic import BaseModel
from typing import Optional
from config import settings
from profiler import generate_profile
from scorer import compute_score
from hallucination_guard import validate_score_grounded

logging.basicConfig(level=getattr(logging, settings.LOG_LEVEL.upper(), logging.INFO))
logger = logging.getLogger(__name__)

# ─── Auth ─────────────────────────────────────────────────────

security = HTTPBearer()

def verify_token(credentials: HTTPAuthorizationCredentials = Depends(security)):
    if credentials.credentials != settings.PYTHON_AI_SERVICE_SECRET:
        raise HTTPException(status_code=401, detail="Invalid authentication")
    return True


# ─── Models ───────────────────────────────────────────────────

class CompanyInput(BaseModel):
    id: Optional[str] = None
    name: str
    industry: str = ""
    employee_count: Optional[int] = None
    description: str = ""
    website_text: str = ""
    linkedin_description: str = ""
    tech_stack: list[str] = []
    ai_job_count: int = 0
    pain_signals: list[str] = []
    service_match: Optional[str] = None

class ContactInput(BaseModel):
    full_name: str = ""
    email: Optional[str] = None
    email_verified: bool = False
    linkedin_personal_url: Optional[str] = None
    social_profiles: dict = {}

class ProfileRequest(BaseModel):
    company: CompanyInput
    contacts: list[ContactInput] = []
    trace_id: str = ""


# ─── App ──────────────────────────────────────────────────────

@asynccontextmanager
async def lifespan(app: FastAPI):
    logger.info("🚀 AI Profiling Service v2 starting...")
    logger.info(f"   NVIDIA NIM: {settings.NVIDIA_NIM_BASE_URL}")
    logger.info(f"   Models: GPT OSS → Gemma 3 → LLaMA 70B → LLaMA 8B → Deterministic")
    yield
    logger.info("AI Profiling Service shutting down")

app = FastAPI(
    title="AI Lead Profiling Service",
    version="2.0.0",
    lifespan=lifespan,
)


# ─── Endpoints ────────────────────────────────────────────────

@app.get("/health")
async def health():
    return {
        "status": "healthy",
        "version": "2.0.0",
        "models": {
            "primary": "nvidia/llama-3.1-nemotron-ultra-253b-v1",
            "secondary": "google/gemma-3-27b-it",
            "tertiary": "meta/llama-3.3-70b-instruct",
            "fast": "meta/llama-3.1-8b-instruct",
        },
    }


@app.post("/profile")
async def profile_company(request: ProfileRequest, _auth: bool = Depends(verify_token)):
    """
    Full profiling pipeline:
    1. LLM generates profile (chain-of-thought, grounded)
    2. LLM extracts signals for scoring
    3. Code computes score deterministically
    4. Both are validated for hallucinations
    """
    company_data = request.company.model_dump()
    contacts_data = [c.model_dump() for c in request.contacts]
    trace_id = request.trace_id

    try:
        # Step 1: Generate profile (LLM with grounding)
        profile = await generate_profile(company_data, trace_id)

        # Step 2: Compute score (LLM extracts signals → code computes)
        score = await compute_score(company_data, profile, contacts_data, trace_id)

        # Step 3: Validate score consistency
        score_validation = validate_score_grounded(score, profile)
        if not score_validation["is_valid"]:
            logger.warning(f"Score validation issues: {score_validation['issues']}")

        return {
            "profile": profile,
            "score": score,
            "validation": {
                "profile_grounded": profile.get("grounding_score", 0),
                "profile_consistent": profile.get("is_consistent", True),
                "score_valid": score_validation["is_valid"],
                "score_issues": score_validation.get("issues", []),
            },
            "meta": {
                "model_used": profile.get("llm_model", "unknown"),
                "is_fallback": profile.get("is_fallback", False),
                "tokens_used": profile.get("tokens_used", 0),
                "trace_id": trace_id,
            },
        }

    except Exception as e:
        logger.error(f"Profiling failed for {company_data.get('name')}: {e}")
        raise HTTPException(status_code=500, detail=str(e))


# ─── Run ──────────────────────────────────────────────────────

if __name__ == "__main__":
    import uvicorn
    uvicorn.run("main:app", host="0.0.0.0", port=8000, reload=False)