Tailored AI Solutions · Use Cases by Industry
Six Industries,
six concrete AI solutions.
No standard products. Every solution is embedded in your Django application and talks directly to your database. Here you can see concretely what that looks like — per industry, with examples and technical details.
● AI solutions
Practices & Secretariats: AI answers the phone.
Hearing aid clinics, medical practices and therapists lose hours every day on calls about appointments, cancellations and standard questions. We build AI agents that answer the phone, talk to your Django or PostgreSQL database and forward only the truly complex cases to the team.
- Voice Agent with Whisper + Llama 3.1 for DE/IT/EN
- Direct connection to the appointment database (availability checked live)
- Patient identification by date of birth + name, GDPR-compliant
- Escalation to the team for emergency keywords or ambiguity
Sample code · Django + Agent
# patient_agent.py — KI nimmt Anruf entgegen from codlab.ki import VoiceAgent from praxis.models import Patient, Termin agent = VoiceAgent( llm = "llama-3.1-70b", asr = "whisper-large-v3", deploy = "on-premise", ) def handle_call(caller): patient = Patient.objects.match( name = caller.spoken_name, dob = caller.spoken_dob, ) if patient.verify(): return agent.book_termin(patient) agent.escalate_to_team()
On-Premise Stack
# docker-compose.yml — alles im Firmennetz services: llama: image: ollama/ollama:latest runtime: nvidia volumes: [./models:/root/.ollama] monitor-agent: image: codlab/monitor-agent:1.0 environment: LLM_URL: http://llama:11434 LOG_SOURCE: /var/log/syslog # Auf eigenem Server # Keine Logs verlassen das Netz pgvector: image: ankane/pgvector:latest
● AI solutions
IT Infrastructure: Llama on your server, not in the cloud.
For companies with their own server infrastructure, we deploy LLMs (Llama 3.1, Mistral, DeepSeek) on local GPU hardware. The agent analyzes logs, classifies incidents, creates tickets, and reports anomalies — without a single byte leaving your firewall.
- Llama 3.1 70B or Mistral Large on RTX 4090 / L40S
- Log classification (critical / warning / noise) in real-time
- Auto-ticket in Jira/Zammad/OTRS with reproduction steps
- Maintenance & updates included — we know Ollama + vLLM ProdOps
RAG · pgvector + Llama
Law Firms: RAG on tens of thousands of PDFs.
Mandates, judgments, contracts — a law firm has tens of thousands of PDFs that no one can fully search. We build a RAG system with local Llama and pgvector that delivers precise answers with source citation in seconds. No client data leaves the firm.
- Ingestion pipeline for PDF, DOCX, scanned documents (OCR)
- Semantic search with source citation and confidence score
- Permissions per mandate — RAG only sees the authorized documents
- BRAO + GDPR-compliant, On-Premise or dedicated EU-VM
RAG-Pipeline · simplified
# Frage stellen, Antwort mit Quelle bekommen from codlab.rag import RAG rag = RAG( llm = "llama-3.1-70b", embed = "bge-m3", store = "pgvector", perms = mandant_filter, ) answer = rag.query( "Klausel Haftungsausschluss BGB §831?" ) for source in answer.sources: print(source.file, source.page) # Mandant-2024-Mueller.pdf p.12 # Konfidenz: 0.94
Multi-Channel Reservation Agent
# gastro_agent.py from codlab.ki import Agent from restaurant.models import Tisch, Reservierung agent = Agent( llm = "claude-haiku", # Cloud reicht: keine sensiblen Daten channels = ["web", "email", "telefon"], ) def handle_reservation(request): tisch = Tisch.objects.available( datum = request.datum, pers = request.personen, ) if tisch: return Reservierung.create(...) agent.suggest_alternative(...)
Cloud API · Claude / GPT-4
Hotel & Gastronomy: Reservation agent across all channels.
Reservations come via web, email, phone, sometimes Instagram. An AI agent unifies these channels, checks table availability live in your database, suggests alternatives, and automatically sends confirmation + reminders. Direct connection to Riservati or your existing system.
- Cloud-LLM (Claude Haiku or GPT-4) — latency more important than On-Premise
- Multilingual DE/IT/EN/FR, with culturally appropriate tone
- Integration with POS, table plan, reservation CRM
- No-show reduction through automatic reminders 24 hours in advance
Workflow agents · Hybrid
Craft & Production: AI extracts orders from emails.
Orders come via email with PDF, photo, sometimes handwritten notes. An AI agent extracts the structured data (items, quantity, delivery date), matches it with your inventory system, and automatically creates the order — or asks for clarification if something is unclear.
- OCR + Vision-LLM (Llama 3.2 Vision or GPT-4 Vision)
- Connection to Sage, Lexware, DATEV or your ERP
- Plausibility check against stock and delivery times
- Human-in-the-loop: In case of uncertainty, ask for clarification, no blind auto-posting
Email-to-Order Pipeline
# auftrag_extractor.py from codlab.ki import VisionAgent from erp.models import Auftrag, Artikel vision = VisionAgent( llm = "llama-3.2-vision", schema = AuftragSchema, ) def on_email(mail): data = vision.extract(mail.attachments) if data.confidence > 0.92: Auftrag.objects.create(**data) else: vision.ask_human(data.unclear)
AI Act risk classification
# Pragmatischer Audit-Workflow 1. Inventar of all AI systems in operation 2. Klassifizierung according to AI Act risk ↳ minimal # Spam-Filter, Suche ↳ limited # Chatbot, RAG, Generator ↳ high # HR, Kredit, biometrisch ↳ verboten # Social-Scoring 3. Dokumentation Obligations per level 4. DSGVO-Cross-Check 5. Roadmap Compliance by 2026/2027 # Liefergegenstand: 8–15 Seiten, # keine 200-Seiten-Kanzleigutachten
EU AI Act · GDPR
Consulting & Audit: AI Act without panic.
You already run AI (even if you may not call it that) and wonder what the EU AI Act will require of you from August 2026. We run a pragmatic audit, classify your systems and deliver a roadmap — not a 200-page legal opinion, but a list of “do this, not that”.
- Inventory of all existing AI systems (internal and external)
- Risk classification according to AI Act (minimal / limited / high)
- Mandatory documentation and transparency requirements
- GDPR cross-check: where do the obligations overlap?
Is your industry missing? Let's talk.
The six solutions shown here are just the most common. We have also equipped cooperatives, cemetery administrations, and furniture manufacturers with AI. Tell us what you want to automate — we will honestly say if AI fits here.