CybersecurityArtificial Intelligence (AI)Desktop DevelopmentSoftware EngineeringData ScienceVerified engineering case

Smart Network Mapper — Plateforme intelligente d’audit réseau

Discover, analyze and prioritize network risks with AI.

Smart Network Mapper is a desktop cybersecurity application that discovers hosts on a network, scans TCP/UDP services, and analyzes potential vulnerabilities. It combines a local Random Forest model, OSINT NVD data, HTML/JSON reports, and Telegram automation via n8n.

View source code Open live demo

01. Context & Challenge

Problem addressed

Local network audits require combining equipment discovery, service identification, version analysis, and vulnerability databases, which is complex and time-consuming to achieve manually.

02. Engineering Goals

Objectives

Automatically detect the local network and its active hosts.

Scan TCP and UDP ports in quick, complete, or custom modes.

Identify services, retrieve their banners, and extract detectable versions.

Predict the vulnerability risk of services using a local model.

Enrich versioned services with CVEs from the NVD.

Store, visualize, and export audit results.

Allow remote control of scans from Telegram.

03. Architecture & System

Architecture

The application combines a CustomTkinter interface and an automatable CLI with a hybrid ARP/TCP discovery engine, a concurrent port scanner, IANA service resolution, and a local AI pipeline. The results are historicized in SQLite, exported in JSON and HTML, and can be supplemented by an audit written by Groq as well as distributed by an n8n/Telegram workflow.

Architectural components & pipeline

›CVE dataset generator: NVD collection, CPE analysis, service/port/protocol association, simulation of fixed versions, and one-hot encoding.

›Training pipeline: RobustScaler normalization, QuantileTransformer transformation, Random Forest training, and export of .pkl artifacts.

›Model distribution: controlled download from Hugging Face on first launch.

›User inputs: Windows GUI with CustomTkinter, interactive CLI, or JSON CLI for n8n.

›LAN detection: identification of active interface and CIDR calculation with psutil.

›Host discovery: ARP broadcast with Scapy, system ARP cache, then concurrent TCP fallback.

›Equipment enrichment: reverse DNS, MAC address, and OS estimation by TTL.

›Network scan: concurrent TCP/UDP sockets, state detection, banner grabbing, and version extraction.

›Service resolution: updated IANA cache every 30 days with static fallback.

›AI engine: port/version/service/protocol vectorization, RobustScaler, QuantileTransformer, Random Forest, and heuristic layer.

›OSINT enrichment: NVD CVE search for services with detected versions.

›Reporting: dashboard, filterable table, SQLite history, JSON, HTML report, and Markdown audit with Groq.

›Automation: n8n workflow with Telegram bot, target selection, CLI execution, and report sending.

›Distribution: PyInstaller for Windows, portable package, and deployable React documentation with Vite/Nginx/GitHub Pages.

04. Metrics & Outcomes

Documented results

The model dataset contains 2,287,965 lines and 91 prediction variables.

The distributed Random Forest model weighs 5,133,005,737 bytes.

Three transformation artifacts accompany the model.

The scanner proposes 24 priority TCP ports.

The scanner proposes 13 priority UDP ports.

The scanner allows a full scan of the 65,535 TCP ports.

Audits are stored in a local SQLite database using WAL mode.

The n8n workflow includes 16 nodes.

The repository contains 52 test functions.

The project provides a portable Windows package.

The project includes a Dockerfile for the Python engine.

The project includes a React/Nginx Dockerfile for the documentation site.

Back to all projects