Rayane
Louzazna.
I wire AI into production systems that already work. No rebuild. No months of dev. Just results.
Projects.
AI-Driven ERP for Affiliate Marketing
5 disconnected platforms, no unified view, 8h manual reporting every week.
Full-stack ERP consolidating ClickUp, Everflow, BuyGoods, Digistore, and ClickBank into one intelligent system with natural language querying, fraud detection, and automated workflows.
- —Text-to-SQL: anyone on the team queries 51 tables in plain English
- —Fraud detection auto-suspends high-risk affiliates via Everflow API
- —Unified data from ClickUp, Everflow, BuyGoods, Digistore, ClickBank
- —Slack AI agent for real-time business queries
8-hour weekly reporting process → under 30 minutes.
ValoML — Esports Scouting Platform
“Manual VALORANT scouting takes hours and misses statistical patterns.”
Production MLOps pipeline analyzing 50 pro matches using K-Means clustering. Full observability stack — MLflow, Prometheus, Grafana. Tactical reports via Llama 3.3.
roadmapAi — Personalized Learning Paths
“Generic learning resources waste months. No AI tool builds a structured path around your exact goal.”
SaaS platform that transforms a learning goal into a structured, week-by-week roadmap. LLM understands your context — stack, level, timeline — and generates a personalized curriculum.
Experience.
5 disconnected platforms (ClickUp, Everflow, BuyGoods, Digistore, ClickBank) — no unified view, 8-hour weekly reporting process.
- —Unified data from 5 platforms into a single ERP — 51 database tables
- —AI interface to query all business data in plain English (Text-to-SQL, 4 LLM providers incl. Claude API)
- —11 automated n8n workflows replacing manual operations tasks
- —Slack AI agent for real-time team queries against live business data
- —Full deployment: Next.js · TypeScript · MySQL · Docker · DigitalOcean
8-hour manual reporting process → under 30 minutes per week.
No structured way for self-learners to get a personalized, actionable learning path.
- —Built and deployed a SaaS platform generating personalized learning roadmaps via LLMs
- —Designed full-stack architecture: Next.js, Supabase, Clerk, Vercel
- —LLM-based feature transforms user goals into structured 8-week step-by-step paths
- —Handled product definition, feature prioritization, and end-to-end deployment solo
Solo built — from idea to deployed SaaS.
Manual esports scouting takes hours and misses statistical patterns in player behavior.
- —K-Means clustering for playstyle analysis — Silhouette Score 0.99, tracked with MLflow
- —Real-time tactical insights via Groq (Llama 3.3 70B) with sub-second inference
- —FastAPI + Next.js, containerized with Docker Compose (5 services)
- —Monitoring via Prometheus + Grafana; integrated GRID Esports API with smart caching
Full scouting reports in ~15 seconds vs hours of manual analysis.
IT consumables managed entirely on paper — untraceable, error-prone process for 500+ staff.
- —Built a fullstack Next.js app (React + API routes) with secure NextAuth authentication
- —Automated PDF generation with electronic signature (jsPDF + React Canvas)
- —Business workflow with 9 statuses and role-based access control per department
- —Modular architecture, secure parameterized queries, versioning via GitHub Actions
Validated and adopted in production at Algeria's largest energy company.
AI Integration Framework
AI works in layers.
Which one do you need?
Most teams think "AI" means a chatbot. The real leverage is in all three levels working together — like at Xentraffic.
Level 1
Embedded Intelligence
Your system thinks, automatically.
Rules that reason. Scoring engines that weigh dozens of signals. Matching algorithms that find the right result without a user telling them what to look for. No LLM — just hard logic done right.
Typical stack
Level 2
Operational Automation
Your system acts, without human input.
The moment a threshold is crossed, something happens — a Slack alert fires, a report lands in an inbox, a high-risk affiliate gets suspended. Workflows replace the human who used to do it manually at 9am.
Typical stack
Level 3
Conversational AI
Your system speaks your business language.
Ask a question in plain English. Get an answer pulled from 51 live database tables. No dashboard, no SQL, no analyst in the loop — just a conversation that routes through the right LLM and returns real data.
Typical stack
At Xentraffic, all three layers run together — algorithms score affiliates, n8n automates the response, and the LLM layer lets the team interrogate everything in plain English.
Discuss my project →Stack.
Ready when
you are.
Available for AI integration, MLOps pipelines, and fraud detection projects. Remote · Worldwide.