I build for production, not for demos. Whether it's a streaming pipeline, an anomaly-detection model, or an LLM extraction service, I focus on systems that are reliable, observable, and honest about their limits - precision over hype, every time.
ACHRAF CHAHBOUNE
Document AI Engineer
French & Arabic Document Extraction
Data Engineer - Kafka, PySpark, Airflow
PhD Candidate - Multi-Agent RL
ACHRAF CHAHBOUNE
Document AI Engineer
French & Arabic Document Extraction
Data Engineer - Kafka, PySpark, Airflow
PhD Candidate - Multi-Agent RL
About Me
I'm an AI engineer. I build the systems that get messy real-world data into production - document extraction, real-time data pipelines, and the models that run on top of them.
Most of my work is document AI and data engineering. I go deeper than most on French and Arabic - right-to-left, cursive ligatures, diacritics, mixed-language pages - and on regulated environments where the data can't simply leave the country. The job is the same either way: ship it, then keep it reliable.
Tell me what you're trying to automate - I'll tell you honestly whether AI is the right tool.
- Based in: Casablanca, Morocco
- Languages: French, Arabic, English
- Freelance: Available
- Focus: Document AI & Data Engineering
My Services
Document AI & Extraction
Turning French and Arabic legal, fiscal, and administrative documents into clean structured data. Vision-language models, continuous fine-tuning (LoRA/PEFT), and human-in-the-loop review, served as FastAPI microservices on GCP.
Data Engineering
Real-time and batch ETL with Kafka, PySpark, and Airflow - cleaned, monitored, and delivered into Power BI. Production pipelines built to scale and fully containerized with Docker.
Applied Machine Learning
Anomaly detection, computer vision, and medical-imaging models with PyTorch and scikit-learn - from data preprocessing through evaluation and deployment.
MLOps & Cloud
Observable, reproducible delivery: Docker, GCP Cloud Run, Prometheus and Grafana monitoring, and CI/CD.
By the Numbers
4M+
Records / day
Records / day
+25%
Incidents caught
Incidents caught
−40%
Inference time
Inference time
How I Work
Work With Me
I take on a small number of engagements at a time, each scoped in writing before any work starts. Remote, in French or English, from Casablanca. If none of these fit, ask anyway.
Document Extraction Pilot
2-3
weeks
- Your real French or Arabic documents, not a demo set
- Per-field accuracy measured and reported honestly
- Go / no-go recommendation, with the cost of scaling it up
Real-Time Data Pipeline
4-8
weeks
- Ingestion and transformation with Kafka, PySpark and Airflow
- Containerised and deployable to your own cloud, monitored from day one
- Runbook and handover so your team owns it, not me
Anomaly Detection & Applied ML
3-6
weeks
- Baseline established on your own historical data
- Tuned to the false-positive rate your team can live with
- Served as a FastAPI service, with a retraining plan
Data & ML Audit
1
week
- A review of your pipeline, models or extraction stack
- What is fragile, what is costing you, what to fix first
- Written findings with prioritised recommendations
How An Engagement Runs
1. A 30-minute call. You describe the problem. I tell you whether I'm the right person for it, and say so plainly if I'm not.
2. A written scope. Deliverables, timeline and price, agreed before any work begins. No open-ended hourly drift.
3. The build, with a weekly check-in. You see progress every week, not a reveal at the end.
4. Handover. Code, documentation and a runbook, so your team can operate and extend it without me.
What I need from you: sample data or access to it, one person who can make decisions, and a clear definition of what 'working' means for your case.
Resume
Experience
Aug 2025 – Present | Casablanca
AI Engineer – Document Intelligence
Simulator Online
- Built the production document-extraction platform for a CNDP-accredited legal and tax SaaS: vision-language models on French and Arabic legal and fiscal documents
- Continuous LoRA/PEFT fine-tuning on a human-in-the-loop review cycle
- FastAPI microservices with PostgreSQL and Docker on GCP Cloud Run
- Observability and monitoring across the full extraction pipeline
Feb 2025 – Aug 2025
Data & AI Engineer
Orange Morocco
- Built real-time telecom pipelines with PySpark and Airflow - 4M+ records/day
- Anomaly detection (Isolation Forest, Autoencoders, time series) - +25% incidents caught before customer impact
- Cut KPI latency from hours to minutes and unified the data lake
- Served Power BI dashboards via REST APIs, with MLOps for retraining
Jul 2024 – Sep 2024 | Hybrid
AI/ML Intern — 3D Medical Imaging
3D Smart Factory
- Built a 3D U-Net for brain-MRI tumor segmentation (T1/T2/FLAIR) - 0.85 Dice
- Cut inference time by 40% via model and pipeline optimization
- Shipped a FastAPI inference service with experiment tracking in MLflow
- Built preprocessing, evaluation, and 3D visualization tooling in PyTorch
Apr 2023 – Aug 2023 | Casablanca, Morocco
Full Stack Developer Intern
BOTTU
- Built a stock-management system with React.js + PHP + MySQL
- Integrated barcode scanning, product categorization, and search
- Streamlined warehouse operations and reduced manual work
Education
2025 – 2028
PhD — Computer Science (Multi-Agent RL)
ENSAM Casablanca · Lab AIMCE
Constrained multi-agent reinforcement learning for autonomous cloud resource management.
2023 – 2025
MSc — Big Data & IoT
ENSAM Casablanca
Data engineering, distributed systems, cloud computing, and applied AI.
2020 – 2023
BSc — Computer Science (IRM)
FSTM Mohammedia
Foundations in databases, algorithms, software engineering, Python, and networking.
Skills
Document AI & LLM
-
Vision-Language Models
-
LoRA / PEFT Fine-Tuning
-
French & Arabic Document Extraction
-
Human-in-the-Loop Review
-
RAG & Retrieval
-
Hugging Face
-
vLLM
Languages
-
French - native
-
Arabic - native
-
English - professional
Data Engineering & Platform
-
Python
-
SQL
-
Apache Kafka
-
PySpark
-
Apache Airflow
-
PostgreSQL / MySQL
-
FastAPI
-
Docker
-
GCP Cloud Run
-
Power BI
Machine Learning & Research
-
PyTorch
-
TensorFlow
-
scikit-learn
-
Anomaly Detection
-
Computer Vision & Medical Imaging
-
MLOps & CI/CD
-
Prometheus / Grafana
-
Multi-Agent Reinforcement Learning
Works
PharmaFind
Software Engineering
DQN Trading Agent
AI & ML
Gaze-Controlled Keyboard
AI & ML
Real-Time Telecom Data Pipeline (Orange Morocco)
Data Engineering
Real-Time ETL Pipeline with Anomaly Detection
Data Engineering
Stock Management System
Software Engineering
Blog
June 30, 2026
Putting LLM Document Extraction into Production
Extracting data from documents with an LLM is easy in a demo and hard in production. Lessons on building auditable, self-improving document-automation systems on GCP.
June 25, 2026
Catching Incidents Before Customers Do: Anomaly Detection at Telecom Scale
How a real-time pipeline processing 4M+ records a day used layered anomaly detection – Isolation Forest, Autoencoders, and time-series methods – to catch 25% more incidents before they reached customers.
June 20, 2026
Building a Real-Time ETL Pipeline with Kafka, PySpark, and Airflow
A walkthrough of a streaming ETL architecture that ingests operational data, scores it for anomalies in real time, and lands clean metrics in Power BI – built with Kafka, PySpark, Airflow, and Prometheus.
Get in Touch
- Based in: Casablanca, Morocco
- Email: [email protected]
- LinkedIn: in/achraf-chahboune
- Freelance: Available
Contact Form