convez.ai
PluginAutomaçãoIntermediário
sweNNN-svg

SAP Ticket Router

sweNNN-svg/sap-ticket-router

Roteador SAP inteligente: rules → TF-IDF → LLM, zero APIs desnecessárias

Ver no GitHub
InicialPluginScore · de 0 a 100

Sistema híbrido de roteamento inteligente de tickets SAP com três camadas de decisão: Rule Engine (sem custo), TF-IDF (offline) e fallback com LLM. Minimiza chamadas desnecessárias à API classificando automaticamente tickets para os módulos SAP corretos (MM, SD, FI/CO, HR, etc).

Instalação

$npx convez-ai sweNNN-svg/sap-ticket-router

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 6 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Automação de suporte SAP com classificação inteligente de tickets
  • Redução de custos em chamadas API usando camadas de fallback eficientes
  • Roteamento de tickets para módulos SAP sem intervenção manual

Veredito

Solução prática e bem arquitetada para ambiente SAP empresarial. Excele em eficiência de custos ao evitar LLM quando regras e TF-IDF resolvem o problema, mas restrição proprietária limita adoção comunitária.

README

A tradução para o português deste README ainda está na fila.

SAP Ticket Router

Hybrid SAP support ticket classification using a three-layer decision system.

How It Works

Ticket Input
     ↓
Layer 1: Rule Engine (TCODE match + keyword rules) → 100% confidence, no API call
     ↓ no match
Layer 2: TF-IDF Model → fast, offline, no API call
     ↓ low confidence
Layer 3: LLM Fallback (Claude Haiku) → returns module + reasoning
     ↓
{"module": "MM", "confidence": "high", "reason": "..."}

Demo

Why Three Layers?

  • Rule Engine: Zero latency, zero cost. Known TCODEs and keywords are routed instantly.
  • TF-IDF: Fast and offline. Handles familiar patterns without API calls.
  • LLM Fallback: Only ambiguous tickets reach Claude. Minimizes API cost.

Example

$ python main.py "yetki hatası alıyorum"
{'method': 'rule_override', 'module': 'Authorization', 'confidence': '100%'}

$ python main.py "fatura kesiminde sorun var"
{'method': 'tfidf', 'module': 'E-solutions', 'confidence': '82.25%'}

$ python main.py "malzeme hareketi sırasında sistem donuyor"
{'method': 'llm', 'module': 'MM', 'confidence': 'high'}

Supported Modules

FI/CO · MM · SD · HR · PP · PM · QM · Basis · Authorization · E-Solutions

Stack

  • Python
  • Anthropic Claude API (claude-haiku-4-5)
  • scikit-learn (TF-IDF + Logistic Regression)
  • YAML rule engine
  • pandas

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

echo "ANTHROPIC_API_KEY=your_key" > .env

python main.py "ticket text here"

Tests

python -m unittest discover -s tests -p "test_*.py"

License

Proprietary - All rights reserved. Viewing is permitted for reference only. No license is granted to use, copy, modify, merge, publish, distribute, sublicense, or sell this software without explicit written permission from Emre Hacımustafaoğlu. For clarity, no license is granted to Atos Bilisim ve Danismanlik A.S. (including parent entities, subsidiaries, affiliates, contractors, and employees) or to organizations developing/operating SAP ticket-routing or substantially similar automated ticket-classification/routing solutions.

Windows Encoding Note

If Turkish characters look broken in Windows terminal output, run:

$env:PYTHONIOENCODING="utf-8"
python main.py "ticket text here"

Background

Built from real-world experience managing 250+ SAP BW/4HANA process chains. Rule engine handles high-confidence cases instantly. TF-IDF covers familiar patterns. LLM handles ambiguous tickets with reasoning — only when necessary.

Roadmap

  • v1.1 — LangSmith Tracing — trace every decision, track latency and cost
  • v1.2 — FastAPI endpoint — HTTP API instead of CLI, POST /predict → JSON
  • v1.3 — Streamlit UI — enter ticket, see which layer responded
  • v1.4 — RAGAS eval — consistency report for LLM decisions
  • v1.5 — Jira / Teams integration — auto-assign ticket to the right team
Ver o original no GitHub

Bom para

Relacionados