Orange

An ns-3-based Open RAN simulation environment with native support for intent-based orchestration.

Resource usage metrics for Server-side WebAssembly services, exposed in a Prometheus-compatible format

Cntains a minimalist code to train and Speech-To-Text models in streaming. It is composed of a training script, configuration files (in hparams/ folder) and the architecture components implementation (in models/ folder

Tool for generating RDFS/OWL ontologies and knowledge graphs. It helps to test AI pipelines by creating synthetic data with reliable knowledge structure and characteristic graph patterns. It also aims to advance the topic of generating realistic RDF knowledge graphs through parametric generation and subgraph matching techniques.

Ontology2Graph generates synthetic knowledge graphs using generative AI and graph fusion, taking an RDFS/OWL ontology as input. Typical applications include testing the robustness of AI pipelines, benchmarking databases, and ensuring the replicability of experiments without sharing sensitive data.

The Self-Supervised Anomaly Detection (ssad) library is a Python framework for self-supervised training of reconstruction-based autoencoder models for anomaly detection.

Gather's the Orange Unified Design System documentation. It contains the exaustive list of all the design tokens described in their latest version, and it contains descriptions of the components.

Evaluation metrics for Sound Source Tracking (SST) systems, and dataset processing tools for LibriJump, a dataset of intermittent and moving speakers.

Source code to reproduce the results listed in “Learning High-Dimensional Parity Functions with Product Networks using Gradient Descent”, accepted at ICML 2026.

The Universal Dependencies PUD corpus also annotated in AMR

Bruno / Hurl / Postman collections for LiveNet

Azure Provider for Crossplane.

FGdiffusion, a method that combines a diffusion model with a flattened graph representation as input and employs ontology-based constraints at inference to generate large-scale, domain relevant knowledge graphs.

Ce projet contient le code Python et les données nécessaires à la reproduction des analyses de l'article "Evaluation de métriques de RAG dans un contexte applicatif : une expérience, ses conclusions et ses limites" (Q. Brabant, EvalLLM 2026).

The Symbal-Latin engine is a generic technology for symbol and graphic gesture recognition developed at Orange Labs. It is based on a detailed modeling of 2D gestures (letters, symbols, command gestures) using graphic primitives and employs an original probabilistic approach of causal generative type.

MMD-based algorithms for sequential change detection

Code, prompts, config and logs associated with the research paper "From Zero-Shot to Reward-Aware: Evaluating Prompting and Memory in LLM-Based Cyber Defense Agents"

TUI presentation tool

Injecting semantic in Streaming Automatic Speech Recognition models

a synthetic dataset of high quality for Abstract Meaning Representation

Configure external DNS servers dynamically from Kubernetes resources

Benchmarking Position Bias in LLM-based Conversational Summarization

Kotlin Multiplatform library implementing W3C VC Data Model, SD-JWT and ISO 18013-5 credentials

This repository aims to provide the necessary tools to reproduce the results of studies concerning factual knowledge updating in LLMs. Those studies were conducted by Hichem Ammar Khodja during his PhD.

LiveNet code samples that are associated with blog articles, technical documentation or tutorials

Code Lab for LiveNet

All Code Labs for Livenet

Code to reproduce the experiments discribed in https://aclanthology.org/2025.acl-long.18/

Zabbix Ansible modules