aza/AzA march 2026 - Kopie (13)/services/event_llm_direct.py

"""Kongress-Suche via OpenAI web search – so simpel wie eine ChatGPT-Anfrage."""
from __future__ import annotations

import json
import os
import re
from dataclasses import dataclass
from datetime import date

from openai import OpenAI

SPECIALTY_DE = {
    "dermatology": "Dermatologie", "general-medicine": "Allgemeinmedizin",
    "internal-medicine": "Innere Medizin", "gynecology": "Gynäkologie",
    "anesthesiology": "Anästhesiologie", "cardiology": "Kardiologie",
    "oncology": "Onkologie", "pediatrics": "Pädiatrie",
    "neurology": "Neurologie", "psychiatry": "Psychiatrie",
    "surgery": "Chirurgie", "ophthalmology": "Ophthalmologie",
    "ent": "HNO", "urology": "Urologie", "orthopedics": "Orthopädie",
    "radiology": "Radiologie", "rheumatology": "Rheumatologie",
    "endocrinology": "Endokrinologie", "gastroenterology": "Gastroenterologie",
    "pulmonology": "Pneumologie", "nephrology": "Nephrologie",
    "infectiology": "Infektiologie", "emergency-medicine": "Notfallmedizin",
    "pathology": "Pathologie", "allergology": "Allergologie",
}

REGION_DE = {"CH": "Schweiz", "EU": "Europa", "WORLD": "weltweit", "US": "USA", "CA": "Kanada"}

COUNTRY_MAP = {
    "schweiz": "CH", "suisse": "CH", "switzerland": "CH",
    "deutschland": "DE", "germany": "DE", "österreich": "AT", "austria": "AT",
    "frankreich": "FR", "france": "FR", "italien": "IT", "italy": "IT",
    "spanien": "ES", "spain": "ES", "grossbritannien": "GB", "uk": "GB",
    "united kingdom": "GB", "griechenland": "GR", "greece": "GR",
    "niederlande": "NL", "netherlands": "NL", "usa": "US", "united states": "US",
    "finnland": "FI", "finland": "FI", "dänemark": "DK", "denmark": "DK",
    "schweden": "SE", "sweden": "SE", "portugal": "PT",
    "belgien": "BE", "belgium": "BE", "china": "CN", "japan": "JP",
}

EU_SET = {"DE","AT","FR","IT","ES","GB","GR","NL","BE","PT","FI","DK","SE","CZ",
          "PL","IE","NO","HU","RO","BG","HR","SK","SI","LT","LV","EE","CY","MT","LU","CH"}


@dataclass
class EventCandidate:
    name: str = ""
    startDate: str | None = None
    endDate: str | None = None
    city: str = ""
    country: str = ""
    urlCandidate: str = ""
    shortDescription: str = ""
    organizer: str = ""
    specialtyTags: list[str] | None = None
    regionTags: list[str] | None = None
    confidence: float = 0.9


def _parse_json(text: str) -> list[dict]:
    """Parse JSON aus LLM-Antwort – robust gegen ```json ... ``` Wrapper."""
    t = text.strip()
    t = re.sub(r"^```[a-zA-Z]*\s*", "", t)
    t = re.sub(r"\s*```\s*$", "", t)
    t = t.strip()
    try:
        obj = json.loads(t)
        if isinstance(obj, dict) and "events" in obj:
            return obj["events"]
        if isinstance(obj, list):
            return obj
    except Exception:
        pass
    m = re.search(r"\{[\s\S]*\}", t)
    if m:
        try:
            obj = json.loads(m.group(0))
            if isinstance(obj, dict) and "events" in obj:
                return obj["events"]
        except Exception:
            pass
    return []


def _norm_country(raw: str) -> str:
    s = raw.strip()
    if len(s) == 2 and s.isalpha():
        return s.upper()
    return COUNTRY_MAP.get(s.lower(), s.upper()[:2] if len(s) >= 2 else s)


def _region_tags(cc: str) -> list[str]:
    tags = set()
    if cc == "CH":
        tags.add("CH")
    if cc in EU_SET:
        tags.add("EU")
    if cc in ("US", "CA"):
        tags.add("US")
    return sorted(tags) or ["EU"]


def _safe_date(v) -> str | None:
    s = str(v or "").strip()
    return s if re.fullmatch(r"\d{4}-\d{2}-\d{2}", s) else None


def query_events_direct(
    specialty: str,
    regions: list[str],
    from_date: date,
    to_date: date,
    lang: str = "de",
    limit: int = 40,
) -> list[EventCandidate]:
    key = os.getenv("OPENAI_API_KEY", "").strip()
    if not key:
        raise RuntimeError("OPENAI_API_KEY nicht gesetzt")

    spec = SPECIALTY_DE.get(specialty, specialty)
    reg = ", ".join(REGION_DE.get(r.upper(), r) for r in regions) or "Europa, Schweiz"
    today = date.today().isoformat()

    prompt = (
        f"Suche im Internet nach den nächsten Kongressen und Weiterbildungen "
        f"für {spec} in {reg} ab heute ({today}) "
        f"bis {to_date.isoformat()}.\n\n"
        f"Liste alle wichtigen Kongresse auf die du findest (mindestens 15). "
        f"Gib die Antwort als JSON zurück:\n"
        '{"events": [{"name": "...", "startDate": "YYYY-MM-DD", '
        '"endDate": "YYYY-MM-DD", "city": "...", "country": "CH", '
        '"url": "...", "description": "...", "organizer": "..."}]}'
    )

    model = os.getenv("EVENT_SEARCH_MODEL", "gpt-4o-mini-search-preview").strip()
    client = OpenAI(api_key=key, timeout=80)
    resp = client.chat.completions.create(
        model=model,
        messages=[
            {"role": "system", "content": "Suche im Internet nach echten Kongressen. Antwort NUR als JSON."},
            {"role": "user", "content": prompt},
        ],
    )
    txt = (resp.choices[0].message.content or "").strip()
    rows = _parse_json(txt)

    out: list[EventCandidate] = []
    for r in rows:
        if not isinstance(r, dict):
            continue
        name = str(r.get("name") or "").strip()
        if not name:
            continue
        cc = _norm_country(str(r.get("country") or ""))
        out.append(EventCandidate(
            name=name,
            startDate=_safe_date(r.get("startDate")),
            endDate=_safe_date(r.get("endDate")) or _safe_date(r.get("startDate")),
            city=str(r.get("city") or "").strip(),
            country=cc,
            urlCandidate=str(r.get("url") or r.get("urlCandidate") or "").strip(),
            shortDescription=str(r.get("description") or r.get("shortDescription") or "").strip()[:600],
            organizer=str(r.get("organizer") or "").strip(),
            specialtyTags=[specialty],
            regionTags=_region_tags(cc),
            confidence=0.9,
        ))
    return out