233 lines
9.6 KiB
Python
233 lines
9.6 KiB
Python
import os
|
||
import json
|
||
import requests
|
||
import sqlite3
|
||
import re
|
||
import datetime
|
||
|
||
# --- Helper: Get Gemini Key ---
|
||
def get_gemini_key():
|
||
candidates = [
|
||
"gemini_api_key.txt", # Current dir
|
||
"/app/gemini_api_key.txt", # Docker default
|
||
os.path.join(os.path.dirname(__file__), "gemini_api_key.txt"), # Script dir
|
||
os.path.join(os.path.dirname(os.path.dirname(__file__)), 'gemini_api_key.txt') # Parent dir
|
||
]
|
||
|
||
for path in candidates:
|
||
if os.path.exists(path):
|
||
try:
|
||
with open(path, 'r') as f:
|
||
return f.read().strip()
|
||
except:
|
||
pass
|
||
|
||
return os.getenv("GEMINI_API_KEY")
|
||
|
||
def get_matrix_context(industry_name, persona_name):
|
||
"""Fetches Pains, Gains and Arguments from CE Database."""
|
||
context = {
|
||
"industry_pains": "",
|
||
"industry_gains": "",
|
||
"persona_description": "",
|
||
"persona_arguments": ""
|
||
}
|
||
db_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), 'companies_v3_fixed_2.db')
|
||
if not os.path.exists(db_path):
|
||
return context
|
||
|
||
try:
|
||
conn = sqlite3.connect(db_path)
|
||
c = conn.cursor()
|
||
|
||
# Get Industry Data
|
||
c.execute('SELECT pains, gains FROM industries WHERE name = ?', (industry_name,))
|
||
ind_res = c.fetchone()
|
||
if ind_res:
|
||
context["industry_pains"], context["industry_gains"] = ind_res
|
||
|
||
# Get Persona Data
|
||
c.execute('SELECT description, convincing_arguments FROM personas WHERE name = ?', (persona_name,))
|
||
per_res = c.fetchone()
|
||
if per_res:
|
||
context["persona_description"], context["persona_arguments"] = per_res
|
||
|
||
conn.close()
|
||
except Exception as e:
|
||
print(f"DB Error in matrix lookup: {e}")
|
||
|
||
return context
|
||
|
||
def get_suggested_date():
|
||
"""Calculates a suggested meeting date (3-4 days in future, avoiding weekends)."""
|
||
now = datetime.datetime.now()
|
||
# Jump 3 days ahead
|
||
suggested = now + datetime.timedelta(days=3)
|
||
# If weekend, move to Monday
|
||
if suggested.weekday() == 5: # Saturday
|
||
suggested += datetime.timedelta(days=2)
|
||
elif suggested.weekday() == 6: # Sunday
|
||
suggested += datetime.timedelta(days=1)
|
||
|
||
days_de = ["Montag", "Dienstag", "Mittwoch", "Donnerstag", "Freitag", "Samstag", "Sonntag"]
|
||
return f"{days_de[suggested.weekday()]}, den {suggested.strftime('%d.%m.')} um 10:00 Uhr"
|
||
|
||
def clean_company_name(name):
|
||
"""Removes legal suffixes like GmbH, AG, etc. for a more personal touch."""
|
||
if not name: return ""
|
||
# Remove common German legal forms
|
||
cleaned = re.sub(r'\s+(GmbH|AG|GmbH\s+&\s+Co\.\s+KG|KG|e\.V\.|e\.K\.|Limited|Ltd|Inc)\.?(?:\s|$)', '', name, flags=re.IGNORECASE)
|
||
return cleaned.strip()
|
||
|
||
def get_qualitative_area_description(area_str):
|
||
"""Converts a string with area information into a qualitative description."""
|
||
nums = re.findall(r'\d+', area_str.replace('.', '').replace(',', ''))
|
||
area_val = int(nums[0]) if nums else 0
|
||
|
||
if area_val >= 10000:
|
||
return "sehr große Flächen"
|
||
if area_val >= 5000:
|
||
return "große Flächen"
|
||
if area_val >= 1000:
|
||
return "mittlere Flächen"
|
||
if area_val > 0:
|
||
return "kleine bis mittlere Flächen"
|
||
return "Ihre Flächen" # Fallback
|
||
|
||
def get_multi_solution_recommendation(area_str, purpose_str):
|
||
"""
|
||
Selects a range of robots based on surface area AND requested purposes.
|
||
"""
|
||
recommendations = []
|
||
purpose_lower = purpose_str.lower()
|
||
|
||
# 1. Cleaning Logic (Area based)
|
||
nums = re.findall(r'\d+', area_str.replace('.', '').replace(',', ''))
|
||
area_val = int(nums[0]) if nums else 0
|
||
|
||
if "reinigung" in purpose_lower:
|
||
if area_val >= 5000 or "über 10.000" in area_str:
|
||
recommendations.append("den Scrubber 75 als industrielles Kraftpaket für Ihre Großflächen")
|
||
elif area_val >= 1000:
|
||
recommendations.append("den Scrubber 50 oder Phantas für eine wendige und gründliche Bodenreinigung")
|
||
else:
|
||
recommendations.append("den Phantas oder Pudu CC1 für eine effiziente Reinigung Ihrer Räumlichkeiten")
|
||
|
||
# 2. Service/Transport Logic
|
||
if any(word in purpose_lower for word in ["servieren", "abräumen", "speisen", "getränke"]):
|
||
recommendations.append("den BellaBot zur Entlastung Ihres Teams beim Transport von Speisen und Getränken")
|
||
|
||
# 3. Marketing/Interaction Logic
|
||
if any(word in purpose_lower for word in ["marketing", "gästebetreuung", "kundenansprache"]):
|
||
recommendations.append("den KettyBot als interaktiven Begleiter für Marketing und Patienteninformation")
|
||
|
||
if not recommendations:
|
||
recommendations.append("unsere wendigen Allrounder wie den Phantas")
|
||
|
||
return {
|
||
"solution_text": " und ".join(recommendations),
|
||
"has_multi": len(recommendations) > 1
|
||
}
|
||
|
||
def generate_email_draft(lead_data, company_data, booking_link="[IHR BUCHUNGSLINK]"):
|
||
"""
|
||
Generates a high-end, personalized sales email using Gemini API and Matrix knowledge.
|
||
"""
|
||
api_key = get_gemini_key()
|
||
if not api_key:
|
||
return "Error: Gemini API Key not found."
|
||
|
||
# Extract Data from Lead Engine
|
||
company_raw = lead_data.get('company_name', 'Interessent')
|
||
company_name = clean_company_name(company_raw)
|
||
contact_name = lead_data.get('contact_name', 'Damen und Herren')
|
||
|
||
# Metadata from Lead
|
||
meta = {}
|
||
if lead_data.get('lead_metadata'):
|
||
try: meta = json.loads(lead_data['lead_metadata'])
|
||
except: pass
|
||
|
||
area = meta.get('area', 'Unbekannte Fläche')
|
||
purpose = meta.get('purpose', 'Reinigung')
|
||
role = meta.get('role', 'Wirtschaftlicher Entscheider')
|
||
salutation = meta.get('salutation', 'Damen und Herren')
|
||
cleaning_functions = meta.get('cleaning_functions', '')
|
||
|
||
# Data from Company Explorer
|
||
ce_summary = company_data.get('research_dossier') or company_data.get('summary', '')
|
||
ce_vertical = company_data.get('industry_ai') or company_data.get('vertical', 'Healthcare')
|
||
ce_opener = company_data.get('ai_opener', '')
|
||
|
||
# Multi-Solution Logic
|
||
solution = get_multi_solution_recommendation(area, purpose)
|
||
qualitative_area = get_qualitative_area_description(area)
|
||
suggested_date = get_suggested_date()
|
||
|
||
# Fetch "Golden Records" from Matrix
|
||
matrix = get_matrix_context(ce_vertical, role)
|
||
|
||
# Prompt Engineering for "Unwiderstehliche E-Mail"
|
||
prompt = f"""
|
||
Du bist ein Senior Sales Executive bei Robo-Planet. Antworte auf eine Anfrage von Tradingtwins.
|
||
Schreibe eine E-Mail auf "Human Expert Level".
|
||
|
||
WICHTIGE IDENTITÄT:
|
||
- Anrede-Form: {salutation} (z.B. Herr, Frau)
|
||
- Name: {contact_name}
|
||
- Firma: {company_name}
|
||
|
||
STRATEGIE:
|
||
- STARTE DIREKT mit dem strategischen Aufhänger aus dem Company Explorer ({ce_opener}). Baue daraus den ersten Absatz.
|
||
- KEIN "mit großem Interesse verfolge ich..." oder ähnliche Phrasen. Das wirkt unnatürlich.
|
||
- Deine Mail reagiert auf die Anfrage zu: {purpose} für {qualitative_area}.
|
||
- Fasse die vorgeschlagene Lösung ({solution['solution_text']}) KOMPAKT zusammen. Wir bieten ein ganzheitliches Entlastungskonzept an, keine Detail-Auflistung von Datenblättern.
|
||
|
||
KONTEXT:
|
||
- Branche: {ce_vertical}
|
||
- Pains aus Matrix: {matrix['industry_pains']}
|
||
- Dossier/Wissen: {ce_summary}
|
||
- Strategischer Aufhänger (CE-Opener): {ce_opener}
|
||
|
||
AUFGABE:
|
||
1. ANREDE: Erzeuge KEINE Anrede (wie "Sehr geehrter..."). Starte direkt mit dem ersten Satz.
|
||
2. EINSTIEG: Nutze den inhaltlichen Kern von: "{ce_opener}".
|
||
3. DER ÜBERGANG: Verknüpfe dies mit der Anfrage zu {purpose}. Erkläre, dass manuelle Prozesse bei {qualitative_area} angesichts der Dokumentationspflichten und des Fachkräftemangels zum Risiko werden.
|
||
4. DIE LÖSUNG: Schlage die Kombination aus {solution['solution_text']} als integriertes Konzept vor, um das Team in Reinigung, Service und Patientenansprache spürbar zu entlasten.
|
||
5. ROI: Sprich kurz die Amortisation (18-24 Monate) an – als Argument für den wirtschaftlichen Entscheider.
|
||
6. CTA: Schließe die E-Mail ab und leite zu den nächsten Schritten über, ohne direkt Termine vorzuschlagen oder nach Links zu fragen. Erzeuge KEINE Schlussformel (wie "Mit freundlichen Grüßen").
|
||
|
||
STIL: Senior, lösungsorientiert, direkt. Keine unnötigen Füllwörter.
|
||
|
||
FORMAT:
|
||
[E-Mail Text]
|
||
"""
|
||
|
||
# Call Gemini API
|
||
url = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent?key={api_key}"
|
||
headers = {'Content-Type': 'application/json'}
|
||
payload = {"contents": [{"parts": [{"text": prompt}]}]}
|
||
|
||
try:
|
||
response = requests.post(url, headers=headers, json=payload)
|
||
response.raise_for_status()
|
||
result = response.json()
|
||
# Remove the placeholder from the LLM-generated text
|
||
cleaned_text = result['candidates'][0]['content']['parts'][0]['text'].replace(booking_link, '').strip()
|
||
return cleaned_text
|
||
except Exception as e:
|
||
return f"Error generating draft: {str(e)}"
|
||
|
||
if __name__ == "__main__":
|
||
# Test Mock
|
||
mock_lead = {
|
||
"company_name": "Klinikum Test",
|
||
"contact_name": "Dr. Müller",
|
||
"lead_metadata": json.dumps({"area": "5000 qm", "purpose": "Desinfektion und Boden", "city": "Berlin"})
|
||
}
|
||
mock_company = {
|
||
"vertical": "Healthcare / Krankenhaus",
|
||
"summary": "Ein großes Klinikum der Maximalversorgung mit Fokus auf Kardiologie."
|
||
}
|
||
print(generate_email_draft(mock_lead, mock_company))
|