<elalaoui.dev />
Let's Talk
ai · cloud · infrastructure · security

The Tech Log & Ideas

Engineering insights on machine learning, cloud orchestration, WAF automation, and edge security. #mlops #devsecops #aiops

Machine Learning / AI June 2026

Network Anomaly Detection: Why Isolation Forest Outperforms Deep Learning

In real-time security monitoring, analyzing millions of log lines requires lightweight, unsupervised algorithms. Unlike deep learning approaches that demand heavy GPU resources, Isolation Forest explicitly isolates anomalies instead of profiling normal data. By combining regex parsing with Pandas dataframes, we mathematically isolate malicious scans and brute-force attempts through low decision-tree depths.

This method achieves sub-millisecond inference per request and integrates seamlessly into lightweight security agents — perfect for edge deployments where latency and compute are critical constraints.

#scikit-learn #anomaly-detection #pandas #edge-ai
Cloud & Security / WAF May 2026

Hardening aaPanel + Cloudflare v4: Silent Edge WAF Automation

Deploying aaPanel is convenient, but exposing it directly without protection is a critical flaw. Using Bash automation scripts combined with Cloudflare's API v4, we can instantly orchestrate network-layer security: enforce strict SSL certificates, enable orange-cloud proxy to mask the origin VPS IP, and inject OWASP-level WAF rules at the edge.

The result is a fully automated, silent deployment where the admin panel remains invisible to scanners, and volumetric attacks are mitigated before reaching the origin server — a must-have for production hosting environments.

#cloudflare #waF #aaPanel #edge-security
Generative AI / LLM April 2026

RAG Pipeline Optimization for Private Cloud Documentation

Retrieval-Augmented Generation (RAG) is a game-changer for internal knowledge bases. In a recent project, we built a RAG pipeline that indexes hundreds of technical documents (security policies, infrastructure runbooks) and serves accurate answers via open-source LLMs (Llama 3, Mistral). The trick: hybrid search combining dense embeddings and keyword BM25.

By using FAISS for vector search and a lightweight reranking step, we reduced hallucination rates by 62% and kept latency under 500ms — fully self-hosted inside a private cloud, no data ever leaving the VPC. This is how enterprises can adopt AI without compromising on compliance.

#RAG #llm #faiss #private-cloud
Hosting / Infrastructure March 2026

Serverless Hosting vs Traditional VPS: Cost & Security Tradeoffs

Choosing between serverless platforms (AWS Lambda, Cloudflare Workers) and traditional VPS hosting affects both budget and security posture. Serverless offers automatic scaling and reduced attack surface (no OS to patch), but cold starts and vendor lock-in can be problematic. VPS gives full control, ideal for custom WAF rules and legacy workloads.

We benchmarked several architectures: for bursty traffic patterns, serverless reduced costs by 70% while maintaining DDoS protection via cloud edge. For steady-state applications with strict compliance needs (PCI-DSS), a hardened VPS with fail2ban, CrowdSec, and automated backups proved more predictable and auditable.

#serverless #vps #cloud-cost #devsecops