Open-Source · Community Driven

AI Experiments for
Telecommunications

A collaborative workspace exploring data-driven decision-making through open-source AI capabilities, open datasets, and cutting-edge ML models — purpose-built for the telecom industry.

20+
Experiments
6
Open Datasets
20
Publications
10+
AI Domains
Revenue Assurance
5G Network Ops
Agentic AI
Sustainability
Security
Service Quality
Smart Grid
Customer Experience

Experiments

Each experiment tackles a real telecom challenge with purpose-built AI models, open data, and reproducible pipelines.

Agentic AI

Agentic Telco Framework

Cloud-native framework for autonomous network management through cooperating AI agents. Implements detection, diagnosis, planning, and resolution without human intervention using custom MCP/ACP protocols.

FastAPI WebSocket MCP ACP
Agentic AI

Autonomous 5G Network

Distributed AI agents with real-time anomaly detection and autonomous remediation for 5G core (AMF, SMF, UPF). Features genuine agent-to-agent communication, Ansible integration, and timeline replay up to 100x speed.

ACP/MCP Ansible Multi-LLM
Network Ops

5G Network Operations

Predicts fault occurrence rates in 5G radio networks by fine-tuning Google T5 models on telecom KPIs correlated with FCC complaints and weather data. Enables proactive fault detection before service impact.

T5 Fine-Tuning Flask Multi-GPU
Network Ops

NOC AI Augmentation

AI-augmented Network Operation Center combining Isolation Forest anomaly detection with LangChain/FAISS vector search and GenAI for multi-source data analysis in 5G core networks. Interactive Dash dashboard.

LangChain FAISS Dash
Revenue Management

Revenue Assurance & Fraud

Balanced Random Forest and Transformer models for predicting fraudulent telecom transactions and identifying revenue anomalies. Trained on synthetic billing data with call duration, roaming, and usage features.

Random Forest Transformers ONNX
Service Quality

Service Assurance NPS

Predicts Net Promoter Score from telecom performance metrics using a 175K-parameter Transformer with multi-head attention. Correlates real FCC complaint data and weather conditions with network metrics.

Transformer NPS MAPE
Customer Experience

Churn Prediction

End-to-end customer churn prediction using Balanced Random Forest and LightGBM. Identifies high-risk customers to enable targeted retention, with Docker-ready deployment and Flask serving.

LightGBM Random Forest Docker
Customer Experience

CRM VoiceBot

AI-powered conversational VoiceBot for telecom customer relationship management. Integrates with Model-as-a-Service platforms for GenAI interaction with OpenShift deployment support.

Flask GenAI OpenShift
Customer Experience

Intent Classification

Unified telco customer intent classification achieving 93.95% accuracy for English and 81.44% for Arabic. Fine-tuned Qwen3-4B with 70+ intent categories, replacing fragmented NLP pipelines.

Qwen3-4B SFT vLLM
Advanced AI

LLM Root Cause Analysis

Multi-source 5G root cause analysis combining Isolation Forest anomaly detection with VectorDB log association and GenAI for explainable diagnostics. Chains classical ML with LLM reasoning.

Isolation Forest VectorDB RAG
LLM / GenAI

Telco Expert Portal

Subject Matter Expert portal with 7+ specialized system prompts (Network, Telco, Cloud, Storage experts), file analysis support, persistent sessions, and real-time vLLM metrics visualization.

Gradio vLLM Embeddings
IT Management

ITSM AI Automation

Integrated IT Service Management combining ServiceNow, Ansible Automation Platform, and OpenShift AI. Automatic incident classification with OpenVINO and automated remediation playbooks.

ServiceNow Ansible OpenVINO
AI-RAN

RAN Energy Efficiency

JAX-based neural networks and Deep Reinforcement Learning for energy-efficient 5G RAN. Achieves 30–35% energy savings through intelligent cell sleep scheduling with sub-5ms inference latency.

JAX/Flax DQN TCN
AI-RAN

PUSCH Neural Receiver

Neural receiver for 5G PUSCH replacing conventional channel estimation with end-to-end deep learning. ResNet + Multi-Head Attention architecture achieving 2–3 dB SNR gain with TensorRT FP16 optimization.

NVIDIA Sionna TensorRT 3GPP
Security

SecOps AI

Network security operations using XGBRegressor for anomaly detection and security metrics prediction. Identifies and prevents threats in telecom networks through latency and behavior analysis.

XGBoost Anomaly Detection
Security

IoT Perimeter Security

ML-based perimeter security for telecom base stations using low-cost IoT sensors (ALS, PIR, temperature/humidity) instead of expensive camera systems. Detects unauthorized access in C-RAN deployments.

IoT Sensors Anomaly Detection
Sustainability

Energy Efficiency

Predicts energy efficiency in telco network infrastructure using linear regression models. Targets green telecom initiatives and power consumption optimization across network elements.

Linear Regression Green Telecom
Smart Infrastructure

AI-Powered SmartGrid

Anomaly detection for power consumption in electrical grids, using UCI household power dataset. Balances power use versus predicted demand for forecasting and proactive grid management.

Anomaly Detection UCI Dataset
Connectivity

Starlink QoE Prediction

Quality of Experience prediction for Starlink using Transformer neural networks. Analyzes altitude, visible satellites, weather, and obstructions to help nomadic users anticipate connectivity quality.

Transformer Flask HuggingFace
Easter Egg

DeepSeek on K8s

Deploy DeepSeek models locally on a single Kubernetes node using Ollama and OpenWebUI. Supports model variants from 1.1GB to 404GB on OpenShift with dual NVIDIA RTX 4090 GPUs.

Ollama OpenShift RTX 4090
Easter EggNew

LLM Inference on Intel Mac with AMD GPU via Vulkan

GPU-accelerated LLM inference on Mac Pro 7,1 (2019) with AMD Radeon Pro Vega II Duo GPUs. Bypasses broken Metal tensor API via Vulkan/MoltenVK to unlock 64 GB HBM2 across 4 GPU dies for local AI serving.

Vulkan llama.cpp Vega II Duo Qwen 3.5

Publications

Articles, talks, and interviews from the team exploring AI applications in telecommunications.

Article

The Busy Mom Syndrome

Fatih E. Nar, Robert Shaw, Taneem Ibrahim

Medium · EnterpriseAI
Read Article
Article

Vibe-r's Guide to Galaxy

Fatih E. Nar, Damien Eversmann, David Kypuros, Sofia Romero

Medium · EnterpriseAI
Read Article
Article

The 12-Factor Agent: Why Agentic AI Patterns Look Suspiciously Familiar

Fatih E. Nar, Murat Parlakisik, Onur Cinar

Medium · EnterpriseAI
Read Article
Article

The Red Queen

Fatih E. Nar, Vincent Caldeira

Medium · EnterpriseAI
Read Article
Article

Authorship Is Dead. Long Live Authority.

Fatih E. Nar

Medium · EnterpriseAI
Read Article
Article

Beyond Chat and Copilots: How Enterprises Will Actually Consume AI Agents

Fatih E. Nar

Medium · EnterpriseAI
Read Article
Article

The Best Choice for AI Inference -> vLLM

Fatih E. Nar, Greg Pereira, Yuan Tang, Robert Shaw, Anish Asthana

Medium · EnterpriseAI
Read Article
Article

The AI Engine Is Ready — But Where's the Rest?

Tushar Katarki, Fatih E. Nar, William Caban

Medium · Open 5G Hypercore
Read Article
Article

Lessons Learned from a Telco MCP Backend Experiments

Ian Hood, Robert Shaw, Fatih E. Nar

Medium · Open 5G Hypercore
Read Article
Article

Satisfaction is All You Need!

Fatih E. Nar, Ian Hood, Ranny Haiby et al.

Medium · Open 5G Hypercore
Read Article
Article

Artificially Intelligent Platform Interface (AI-PI)

Fatih E. Nar, Ian Hood, Shujaur Mufti et al.

Medium · Open 5G Hypercore
Read Article
Article

TrueAI4Telco

Azhar Sayeed, Fatih E. Nar et al.

Medium · Open 5G Hypercore
Read Article
Article

AI Accelerators: Performance vs Sustainability

Fatih E. Nar

Medium · Open 5G Hypercore
Read Article
Article

Avoid AI Blindness for Business Vision

Arun Thomas, Fatih E. Nar

Medium · Open 5G Hypercore
Read Article
Video

AI for Network Scalability

Fatih E. Nar

YouTube Interview
Watch
Video

Integrating Gen AI in Networks

Fatih E. Nar

Vimeo Panel
Watch
Article

GenAI: A Game Changer for Telcos

Vinodhkumar Raghunathan, Fatih E. Nar

Fierce Network
Read Article
Article

How to Train a BERT ML Model on OpenShift AI

Alessandro Arrichiello

Red Hat Developers
Read Article
Article

Transforming ITSM with Ansible Automation

Alessandro Arrichiello

Red Hat Developers
Read Article

HuggingFace Collection

Open datasets and trained models ready for experimentation, fine-tuning, and deployment.

About Telco-AIX

An open, collaborative workspace for the future of intelligent networking.

Our Mission

Telco-AIX is a collaborative experimental workspace where we explore data-driven decision-making through open-source AI capabilities and open datasets. Every experiment is designed to tackle real-world telecom challenges — from predicting customer churn to autonomously managing 5G networks — with reproducible code, open data, and transparent methodology.

How It Works

Each experiment is self-contained with its own dependencies, data pipelines, and deployment patterns. We use a broad range of AI techniques — classical ML (Random Forests, XGBoost), deep learning (Transformers, DQN), LLM-powered RAG pipelines, and multi-agent frameworks with custom protocols (MCP/ACP). Models and datasets are published openly on HuggingFace for the community to build upon.

Key Contributors

Great minds & talent from industry leaders and a growing open-source community.

Red Hat AWS Google Azure Verizon AT&T NVIDIA AMD Intel and others

See our contributors in the repo.

Join the AI Revolution in Telecom

Interested in cutting-edge AI applications for telecom? Let's collaborate.

Connect on LinkedIn