AI / ML ENGINEER · BUILDER OF PRODUCTION AI

I build AI systemsthat survive production.

Multi-agent systems — planned DAGs, not prompt chains.Hybrid RAG — retrieval that grounds, cites, and refuses.Pipelines that handle corrupted input and replan on failure.Every system shipped with a real, passing test suite.

Casablanca → the world

THE PHILOSOPHY

Models hallucinate. Systems shouldn’t.

The model is the easy part. I build what runs around it — pipelines that handle corrupted input, retrieval that admits uncertainty, agents that replan when they fail. Every system I ship is designed around what happens when things go wrong, because in production, they always do.

the arcAURA proved I can wrangle any data format. nexus proved I can build production memory systems. MAESTRO proves I can architect multi-agent AI at depth.

Selected work

Three production systems. Live, tested, documented.

Each one is deployed and clickable. The visual in every card is a live, working illustration — and one click opens the real app.

Read these as
The same three systems, described for whoever is reading.

Universal Data Engine

AURA

Ingest any file → clean → explore → AI chat → branded report. Five minutes, not five hours.

Ingest any fileClean · 8 stepsExploreGrounded AI
any file → insight
Live demoOpen live

A full-stack data intelligence platform. Drop any CSV, XLSX, JSON or Parquet and AURA infers what each column actually is — identifier, measure, dimension, temporal — runs an 8-step cleaning pipeline, builds an archetype-aware explore, answers grounded questions in EN/FR, and exports a fully bilingual (EN/FR) branded PDF report — labels, AI executive summary, insights, charts, and number/date formatting all follow the locale.

Production RAG Engine

nexus

Upload any of 15 formats — instant AI summary, then grounded answers with claim-level source verification.

Auto-Summary
Document uploaded
LiveOpen live

A production-grade Retrieval-Augmented Generation system supporting 15 document formats — PDF, DOCX, XLSX, PPTX, CSV, and more. On upload, an AI auto-summary fires instantly: a one-liner, 4 insight bullets, and 3 clickable question chips that fire directly into chat. Hybrid BM25 + pgvector retrieval with RRF reranking, claim-level answer grounding, contradiction detection and confidence gating — every fact is traceable to a source, and low-confidence answers refuse instead of hallucinating.

Multi-Agent Command Center

MAESTRO

A mission in, a verifiable deliverable out — orchestrated live.

MissionScan the EV market and draft an outreach plan.
ResearchTavily · 8 sources
DataCSV · live stats
Automationn8n · workflow
LiveOpen live

A living neural orchestra. An LLM orchestrator reads a mission, plans a DAG of specialist agents — Research, Data, Automation — runs them as a live SSE pipeline streaming every step, and synthesizes one verifiable deliverable: sourced market briefs, real CSV statistics, importable n8n workflow JSON.

More projects

Beyond the flagships — data, ML, BI.

Projects that round out the picture: a real-time streaming digital twin, time-series forecasting, risk modelling and business intelligence.

Streaming · Forecasting · Evaluated AI

UrbanSense — Casablanca Digital Twin

Final-year team project (4 engineers) on Casablanca’s real road network (OSM), with METR-LA traffic dynamics replayed and labelled, and live TomTom ingestion as a second source: Kafka + Spark Structured Streaming into InfluxDB/MongoDB, a GRU seq2seq forecaster with split-conformal uncertainty (88.5% coverage at 90% nominal), a false-data-injection detector with its blind spot published, and a bilingual assistant evaluated on a 34-question golden set that rejected two LLMs for inventing numbers. I owned the Next.js + deck.gl console and the assistant.

p95 870 ms · MAE 3.07 · 88.5% coverage
KafkaSparkPyTorchNext.jsdeck.gl

Time-Series BI + ML

Walmart Sales Forecasting

End-to-end weekly-sales forecasting across 45 stores — five models benchmarked (Prophet, Prophet + regressors, XGBoost with and without feature engineering, SARIMAX), Optuna-tuned with TimeSeriesSplit CV, a 4-page Power BI report and a 5-page Streamlit app.

2.17% MAPE · Prophet
ProphetXGBoostOptunaPower BIStreamlit

Production ML · Calibrated PD

Credit Risk Prediction

Default-risk scoring on 307K applicants — LightGBM on 153 features (31 engineered from the bureau / credit-history tables), SHAP explainability, isotonic-calibrated probabilities, Gini/KS metrics, and an interactive threshold simulator. Config-driven CLI pipeline with 19 pytest tests. Not a notebook: a validated, calibrated scoring engine.

0.775 AUC · Gini 0.550 · calibrated Brier 0.067
LightGBMSHAPOptunascikit-learnStreamlit

The journey

A path forged through the detour.

Physics, then medicine in a war zone, then a hard pivot into computer science. Each detour added a layer the straight path never could.

2020

Baccalauréat — Physical Sciences

Boujaad, Morocco

Foundation in physics and mathematics.

2021

Russian language year

Ukraine

Preparatory year abroad — a new language, a new continent.

2021 – 2022

General Medicine (English)

Ukraine

Studied medicine — interrupted and ended by the war.

2022

Return to Morocco — the pivot

Casablanca

Left medicine behind and chose technology. The real start.

2022 – 2025

Licence — Computer Science

Mundiapolis · Cloud Computing & Security

Bachelor’s in CS, specialising in cloud computing and security.

2025

AI & Cloud Engineering Intern

B-AGILE, Casablanca

3 months (Jun–Aug): built and deployed a cloud-native AI microservice on Google Cloud — Python, Docker, CI/CD.

2026 – 2027

Engineering — final year

ISGA, Casablanca

UrbanSense, the real-time digital twin, defended Sept 2026 (jury: “excellent work”). 6-month PFE from early February 2027.

2027

Ingénieur d’État — AI & Big Data

Graduating

State engineering degree in Artificial Intelligence & Big Data.

The stack

The tools that ship real systems.

From the AI layer down to cloud, MLOps and data engineering — the stack hiring teams actually look for.

Languages

PythonTypeScriptSQLJavaScriptBash

AI · LLMs · Agents

GroqGeminiClaude APIOpenAI APILangChainRAGMulti-AgentHugging Face

ML · Data Science

PyTorchTensorFlowscikit-learnPandasNumPyXGBoostLightGBMSHAP

MLOps · Pipelines

MLflowAirflowDockerKubernetesCI/CD

Cloud · DevOps

GCPAWSGitHub ActionsVercelHugging Face Spaces

Data · Vectors

PostgreSQLSparkKafkaSupabasepgvectorMongoDBInfluxDB

Backend · Frontend

FastAPINode.jsNext.jsReactTailwind

Design · UI/UX

FigmaUI / UXDesign SystemsFramerWireframingAccessibilityResponsive

Contact

Let’s build something that actually ships.

Looking for a 6-month end-of-study internship (PFE) from early February 2027, then AI / ML engineering roles. Based in Casablanca, open to relocation in Morocco and abroad.