YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
TelecomGPT-Arabic
TelecomGPT-Arabic is a multi-agent Arabic telecom AI system designed to understand customer complaints, support technical troubleshooting, classify telecom issues, and assist with ticket preparation.
The project is built in collaboration with du and ITU to explore how Arabic-focused AI can support real telecom customer-care workflows, especially in scenarios involving Arabic dialects, mixed Arabic-English expressions, and operator-specific support processes.
Status: The model checkpoint will be uploaded soon.
This repository currently provides an overview of the system and its architecture.
For more information contact: brahim.mefgouda@ku.ac.ae LinkedIn: https://www.linkedin.com/in/brahimmefgouda/
Overview
TelecomGPT-Arabic is designed as an agentic telecom AI system, not only as a standard chatbot. It combines Arabic language understanding, telecom-domain reasoning, retrieval-augmented generation, and multi-agent workflow automation.
The system focuses on real telecom support scenarios where customers may describe their problems using informal Arabic, dialects, incomplete information, or mixed Arabic-English terms. TelecomGPT-Arabic aims to understand the complaint, classify the issue, ask the right follow-up questions, retrieve relevant knowledge, and generate structured responses or ticket summaries.
Multi-Agent Architecture
TelecomGPT-Arabic follows a multi-agent design, where each agent performs a specific role in the customer-support workflow. The goal is to move from an unstructured customer complaint to a clear diagnosis, grounded response, and structured ticket or escalation summary.
1. Greeting and Interaction Agent
Starts the conversation, welcomes the customer, and guides the interaction in a natural Arabic style. This agent supports a smooth customer experience, especially when the complaint is written in informal Arabic, dialect, or mixed Arabic-English expressions.
2. Issue Understanding Agent
Analyzes the customer complaint and extracts the main problem, such as a network issue, device issue, SIM problem, billing concern, service activation issue, coverage problem, or connectivity degradation. This agent identifies the customer intent and checks whether the complaint is complete or still missing important information.
3. Classification Agent
Classifies the case into the correct telecom category before sending it to the appropriate workflow. This helps reduce wrong troubleshooting steps and improves the consistency of the support process. For example, the system can distinguish between device-related issues, network-related issues, service-activation problems, billing cases, and cases that require escalation.
4. Multi-Turn Diagnostic Agent
Handles the conversation in a multi-turn manner. If the customer complaint is unclear or incomplete, this agent continues asking focused follow-up questions until the issue is sufficiently understood.
Instead of asking generic or repeated questions, the agent asks only for the missing information needed to diagnose the case, such as the location, device type, SIM status, network mode, error message, affected service, time of the issue, and whether the problem happens indoors, outdoors, or in a specific area.
5. RAG Knowledge and Network Exploration Agent
Retrieves relevant telecom knowledge, FAQs, policies, troubleshooting steps, and operator-specific guidance before generating the final response.
If the required information is not available in the knowledge base, the agent does not guess. Instead, it can explore additional available sources, such as network-status information, coverage indicators, service availability, outage reports, base-station or area-level information, and previous similar cases.
If the issue still cannot be resolved confidently, the agent prepares the case for escalation with a clear explanation of what information is missing and what should be checked next.
6. Response Generation Agent
Generates a clear, customer-friendly response based on the collected information and retrieved knowledge. The response is adapted to the customer language and issue type, while keeping the answer grounded in available telecom knowledge and avoiding unsupported claims.
7. Ticketing and Escalation Agent
Generates a structured ticket summary that includes the customer issue, collected diagnostic information, classification label, retrieved evidence, recommended action, and escalation notes. This helps support agents understand the case quickly and reduces the time needed to open, update, or escalate tickets.
Intended Use
TelecomGPT-Arabic is intended for research, demonstration, and telecom AI development in areas such as:
- Arabic telecom customer support
- AI-assisted call-center workflows
- Telecom complaint classification
- Technical troubleshooting
- Operator-grounded RAG systems
- Multi-agent customer-care automation
- Arabic LLM adaptation for telecom applications
Collaboration Context
TelecomGPT-Arabic is being developed in collaboration with du and ITU.
The goal is to demonstrate how Arabic telecom-focused AI systems can support operator workflows, improve customer interaction, and enable more intelligent digital transformation in telecom services.