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.
Each experiment tackles a real telecom challenge with purpose-built AI models, open data, and reproducible pipelines.
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.
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.
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.
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.
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.
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.
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.
AI-powered conversational VoiceBot for telecom customer relationship management. Integrates with Model-as-a-Service platforms for GenAI interaction with OpenShift deployment support.
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.
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.
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.
Integrated IT Service Management combining ServiceNow, Ansible Automation Platform, and OpenShift AI. Automatic incident classification with OpenVINO and automated remediation playbooks.
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.
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.
Network security operations using XGBRegressor for anomaly detection and security metrics prediction. Identifies and prevents threats in telecom networks through latency and behavior analysis.
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.
Predicts energy efficiency in telco network infrastructure using linear regression models. Targets green telecom initiatives and power consumption optimization across network elements.
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.
Quality of Experience prediction for Starlink using Transformer neural networks. Analyzes altitude, visible satellites, weather, and obstructions to help nomadic users anticipate connectivity quality.
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.
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.
Articles, talks, and interviews from the team exploring AI applications in telecommunications.
Open datasets and trained models ready for experimentation, fine-tuning, and deployment.
Service assurance dataset with ~650K rows of telecom performance metrics, NPS scores, and correlated FCC/weather data.
Revenue assurance dataset with ~1M rows of telecom transaction records including billing patterns, usage, and fraud labels.
5G network operations dataset with ~100K rows of radio network KPIs and fault occurrence records.
Security operations dataset with ~100K rows of network security events, latency metrics, and anomaly indicators.
IoT perimeter security dataset with ~134K rows of sensor readings (ALS, PIR, temperature, humidity) from base stations.
IT Service Management tickets dataset with ~1K categorized incident records for classification training.
An open, collaborative workspace for the future of intelligent networking.
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.
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.
Great minds & talent from industry leaders and a growing open-source community.
See our contributors in the repo.
Interested in cutting-edge AI applications for telecom? Let's collaborate.
Connect on LinkedIn