Skip to content

CLAUDE.md

Latest commit

 

History

History
85 lines (61 loc) · 3.54 KB

CLAUDE.md

File metadata and controls

85 lines (61 loc) · 3.54 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Repository Overview

Telco-AIX is a collection of independent AI/ML experiment projects for the telecommunications domain. Each subdirectory is a self-contained project with its own dependencies — there is no monorepo build system, shared library, or CI/CD pipeline.

Models and datasets are published to HuggingFace: huggingface.co/collections/fenar/telco-aix-66737384ab5687fe3d9a4b94

Running Projects

Each project is independent. The general pattern is:

cd <project-dir>
pip install -r requirements.txt
python main.py  # or the project's entry script

agentic/ and autonet/ (multi-agent frameworks):

pip install -r requirements.txt
python main.py          # starts MCP (port 8000), ACP broker (8002), dashboard (8080)
python main.py --run-test  # agentic: runs with test scenario

telco-sme/ (Gradio UI):

pip install -r requirements-v2.txt
python sme-web-ui-v2.py

5gprod/ (Dash dashboard), crm/ (Flask), churn/ (Flask server on port 5000), revenueassurance/ (Flask): each has a main Python script that starts a web server.

Projects with Dockerfiles: crm/, churn/, etc/model-as-a-server/

Testing

Only agentic/ and autonet/ have pytest tests:

cd agentic && pytest
cd autonet && pytest

No global test runner or linting configuration exists.

Architecture

Protocol-Based Agent Frameworks (agentic/, autonet/)

These are the most complex projects, implementing a multi-agent architecture with two custom protocols:

  • MCP (Model Context Protocol): HTTP/FastAPI-based context and data sharing between agents
  • ACP (Agent Communication Protocol): WebSocket-based inter-agent messaging with a central broker

Four agent types follow a workflow pipeline: Diagnostic -> Planning -> Execution -> Validation. Each inherits from a base Agent class with a state machine (INITIALIZING -> IDLE -> PROCESSING -> WAITING -> TERMINATED/ERROR). The orchestration/service.py coordinates agent workflows.

autonet/ extends agentic/ with real NOC integration, Ansible playbook execution, and multi-backend LLM support (Anthropic, OpenAI, HuggingFace, local) via protocols/mcp/backends/.

Common Tech Stack Across Projects

  • Web UIs: Flask, Gradio, Dash, FastAPI (varies by project)
  • ML/DL: PyTorch, TensorFlow/Keras, scikit-learn, HuggingFace Transformers
  • LLM/RAG: LangChain, OpenAI API, FAISS vector search
  • Model Export: ONNX runtime for inference optimization
  • Data: pandas, numpy; visualization with matplotlib/seaborn/plotly
  • Async: asyncio, websockets, aiohttp (in agent frameworks)
  • Validation: Pydantic (in agent frameworks)
  • Config: python-dotenv for environment variables

Project Domains

Project What It Does
agentic/, autonet/ Multi-agent telco network automation (MCP/ACP protocols)
5gnetops/ 5G fault prediction with BERT/MoE models
5gprod/ NOC dashboard with LLM-augmented operations
telco-sme/ Telco knowledge portal with embeddings search
airan-energy/ AI-RAN energy optimization (DQN + traffic forecasting)
intclass/ Intent classification with fine-tuned Qwen models
llm-rca/ Root cause analysis with LLM chaining and RAG
churn/ Customer churn prediction pipeline
revenueassurance/ Revenue assurance with Random Forest and Transformer models
secops/, iot-sec/ Network and IoT security ML models