AYI NEDJIMI
Senior Offensive Cybersecurity & AI Consultant
Building the bridge between offensive cybersecurity and artificial intelligence. Fine-tuned models, 85+ bilingual datasets, and 40 interactive tools for the security community.
Portfolio Statistics
Live data from Hugging Face API
Last updated: 2026-11-05 21:48 UTC
Fine-Tuned Models
Custom cybersecurity & compliance LLMs trained with QLoRA
🤖 m365-expert-v3GGUF Ready
Base: Qwen/Qwen3-8B | QLoRA Fine-tuning | FR / EN
Expert-level Microsoft 365 administration assistant covering PowerShell, Microsoft Graph, Entra ID, SharePoint, Exchange Online, Intune, and Teams. Full GGUF quantized versions available for Ollama deployment.
🤖 CyberSec-Assistant-3B
Base: Qwen/Qwen2.5-3B-Instruct | QLoRA + LoRA Fine-tuning | FR / EN
Cybersecurity compliance assistant covering GDPR/RGPD, NIS2, DORA, AI Act, ISO 27001, MITRE ATT&CK, OWASP, pentesting, SOC operations, Zero Trust, and DevSecOps.
🤖 ISO27001-Expert-1.5B
Base: Qwen/Qwen2.5-1.5B-Instruct | QLoRA Fine-tuning | FR / EN
Specialized ISO/IEC 27001:2022 expert for information security management systems (ISMS), risk management, security controls, and audit procedures.
🤖 RGPD-Expert-1.5B
Base: Qwen/Qwen2.5-1.5B-Instruct | QLoRA + LoRA Fine-tuning | FR / EN
GDPR/RGPD data protection expert covering DPO, DPIA, consent management, breach notification procedures, and privacy compliance frameworks.
Interactive Spaces
40 Gradio-powered tools, explorers, and demos
🔍 Security Explorers (8 spaces)
🛠 Security Tools (10 spaces)
⚖ Compliance & GRC (9 spaces)
🧠 AI / ML Tools (8 spaces)
Cybersecurity Datasets
85 bilingual (FR/EN) datasets covering the full cybersecurity spectrum
🏛 Standards & Frameworks
8 datasets⚔ Offensive Security
8 datasets🇪🇺 EU Regulations & Compliance
8 datasets🚨 Incident Response & DFIR
6 datasets☁ Cloud & Infrastructure Security
6 datasets🔎 SOC & Threat Intelligence
4 datasets🔮 Emerging Technologies
4 datasets🔗 Supply Chain & SBOM
4 datasetsð M365 & Microsoft Admin
3 datasetsEnterprise Use Cases
Real-world applications of cybersecurity AI models and datasets
SOC Automation
Using CyberSec-Assistant for alert triage
SOC teams are overwhelmed with thousands of daily alerts, leading to analyst fatigue and missed critical incidents. Manual triage is slow and inconsistent across shifts.
Deploy CyberSec-Assistant-3B as an AI co-pilot for L1/L2 analysts. The model classifies alerts using MITRE ATT&CK mapping, suggests investigation steps, and prioritizes based on contextual risk scoring.
70% reduction in mean-time-to-triage. Analysts focus on high-fidelity alerts while the model handles initial classification and enrichment of routine events.
Compliance Assessment
ISO 27001 / RGPD gap analysis
Organizations spend months and significant consulting fees conducting compliance gap analyses across ISO 27001 and GDPR/RGPD frameworks, often with inconsistent results.
Leverage ISO27001-Expert-1.5B and RGPD-Expert-1.5B to automate initial gap assessments. The models analyze existing policies against control requirements and generate detailed findings reports.
Assessment time reduced from weeks to hours. Consistent, bilingual (FR/EN) reports with actionable remediation roadmaps aligned to both ISO 27001:2022 Annex A and GDPR Articles.
Security Training
Interactive cybersec Q&A for teams
Traditional security awareness training is passive, outdated, and fails to engage technical teams. Employees forget content within weeks, leaving organizations vulnerable.
Use the CyberSec-Assistant with RAG pipelines connected to internal policy documents. Teams interact with an AI tutor that answers questions about real attack scenarios, compliance requirements, and company-specific security procedures.
3x improvement in knowledge retention scores. Teams can query security topics on-demand, reducing reliance on scheduled training sessions and empowering continuous learning.
Threat Intelligence
Automated IoC analysis
Threat intelligence teams manually correlate Indicators of Compromise (IoCs) across multiple feeds, taking hours to produce actionable intelligence reports for incident responders.
Combine CyberSec-Assistant with the MITRE ATT&CK and threat hunting datasets to build an automated IoC analysis pipeline. The system maps indicators to TTPs, identifies campaign patterns, and generates structured threat reports.
IoC analysis time cut by 80%. Automated MITRE ATT&CK mapping produces consistent threat reports that integrate directly into SIEM playbooks and incident response workflows.
Downloads & Metrics
Analytics powered by live Hugging Face API data
What Professionals Say
Example use case scenarios illustrating potential enterprise applications
Note: These are illustrative example scenarios, not actual client testimonials.
We integrated the CyberSec-Assistant into our SOC workflow and saw immediate improvements in alert triage consistency. The bilingual capability was a game-changer for our Paris and London teams.
The ISO 27001 and RGPD expert models reduced our compliance assessment preparation time dramatically. The quality of gap analysis outputs rivals what we get from Big 4 consultancies.
Having 85+ bilingual cybersecurity datasets in one place transformed our internal AI training program. We built a custom RAG system for our security team in days instead of months.
The GGUF quantized models running locally via Ollama gave us the air-gapped AI assistant our classified environment needed. No cloud dependency, full control, exceptional quality.
Project Timeline
Key milestones in the CyberSec AI journey
Project Genesis
Started building bilingual cybersecurity datasets covering MITRE ATT&CK, OWASP, and ISO 27001 frameworks.
Dataset Collection Grows to 40+
Published FR/EN datasets for EU regulations (NIS2, DORA, GDPR, AI Act), incident response, forensics, and cloud security.
First Fine-Tuned Models Released
Launched CyberSec-Assistant-3B, ISO27001-Expert-1.5B, and RGPD-Expert-1.5B using QLoRA fine-tuning on Qwen base models.
Interactive Spaces Ecosystem
Built 20+ Gradio-powered security tools including CVE Lookup, Attack Path Visualizer, and Compliance Checker.
85 Datasets Milestone
Reached 85 published datasets spanning the full cybersecurity spectrum. Added supply chain, SBOM, and post-quantum cryptography coverage.
M365-Expert-v3 with GGUF
Released the m365-expert-v3 model with full GGUF quantization support for Ollama local deployment.
40 Spaces & Community Growth
Expanded to 40 interactive Spaces, published 10+ research articles, and grew community engagement across the platform.
Enhanced Portfolio & Metrics
Launched the enhanced portfolio with live metrics, use cases, Plotly analytics, and expanded GGUF/Ollama deployment guides.
GGUF & Ollama Deployment
Run cybersecurity AI models locally with one command -- no cloud, no API keys, full privacy
📦 What is GGUF?
GGUF (GPT-Generated Unified Format) is a quantization format that compresses large language models to run efficiently on consumer hardware. It enables CPU and GPU inference without requiring expensive cloud infrastructure. Our GGUF models are available in multiple quantization levels (Q4_K_M, Q5_K_M, Q8_0) to balance quality vs. speed for your hardware.
🚀 Quick Start with Ollama
curl -fsSL https://ollama.com/install.sh | sh
cat << 'EOF' > Modelfile
FROM ./m365-expert-v3-Q4_K_M.gguf
TEMPLATE """{{ if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER stop "<|im_end|>"
EOF
ollama create m365-expert -f Modelfile
ollama run m365-expert
huggingface-cli download AYI-NEDJIMI/m365-expert-v3 --include "*.gguf" --local-dir ./models
📋 Available GGUF Quantizations
| Quantization | Size (approx) | Quality | Best For |
|---|---|---|---|
| Q4_K_M | ~4.9 GB | Good | Laptops, low-RAM systems |
| Q5_K_M | ~5.7 GB | Very Good | Balanced quality/speed |
| Q8_0 | ~8.5 GB | Excellent | Maximum quality, 16GB+ RAM |
🐍 Use with Python (llama-cpp-python)
pip install llama-cpp-python
from llama_cpp import Llama
llm = Llama(
model_path="./models/m365-expert-v3-Q5_K_M.gguf",
n_ctx=4096,
n_gpu_layers=-1 # Use GPU if available
)
response = llm.create_chat_completion(
messages=[{
"role": "user",
"content": "How to configure Conditional Access in Entra ID?"
}]
)
print(response["choices"][0]["message"]["content"])
Published Articles
Research posts, analysis, and community contributions
About Me
👤 Professional Profile
Senior Offensive Cybersecurity & AI Consultant with deep expertise in penetration testing, red team operations, compliance frameworks, and applied artificial intelligence. Specializing in bridging the gap between advanced cybersecurity practices and cutting-edge AI/ML technologies to build smarter, automated security solutions.
Creator of the largest bilingual (French/English) cybersecurity dataset collection on Hugging Face, with 85+ datasets spanning MITRE ATT&CK, ISO 27001, OWASP, NIST CSF, EU regulations (NIS2, DORA, GDPR, AI Act), incident response, forensics, cloud security, and more. Builder of 4 fine-tuned cybersecurity LLMs and 40 interactive Gradio-powered security tools.
⚔ Offensive Security
- Penetration Testing (Web, Network, AD)
- Red Team Operations
- Active Directory Attacks
- OWASP / MITRE ATT&CK
- Bug Bounty Hunting
- EDR Evasion & Bypass
🤖 AI & Machine Learning
- LLM Fine-tuning (QLoRA, LoRA)
- RAG Pipelines & LangChain
- Prompt Engineering
- MLOps & Model Deployment
- AI Security (LLM Top 10)
- Dataset Engineering
📜 Compliance & GRC
- ISO 27001 / 27002
- GDPR / RGPD
- NIS2 Directive
- DORA Regulation
- EU AI Act
- NIST CSF / CIS Controls
🏆 Key Competencies & Frameworks
🌐 Connect & Follow
AYI NEDJIMI · Senior Offensive Cybersecurity & AI Consultant
Built with Gradio on Hugging Face Spaces