About
Engineering AI systems that sustain themselves in production.

I started as a full-stack and mobile developer in 2018, building apps and APIs for startups across the MENA region. That foundation — owning a system end to end — is still how I approach AI: data pipelines, model training, serving, evaluation and monitoring are one product, not separate hand-offs.
Since 2021 I've worked in machine learning: predictive models and data pipelines as a contractor, then automotive valuation and VIN recognition at eData in Dubai, and now enterprise LLM, agent and document-AI systems for insurance at First.tech Group in Abu Dhabi. Outside work I build and release open models and tools — most recently Jev-AR, an Arabic-first routing model.
- Agentic systems that scale beyond demos
- Building models, not just calling APIs
- Production reliability over prototypes
- Measured performance over default configurations
Experience
Senior AI/ML Engineer · First.tech Group
Jun 2025 – Present
Abu Dhabi, UAE
Building and deploying production AI systems for insurance workflows across LLM applications, agentic automation, document intelligence and multimodal inference.
- Designed a multi-agent workflow for corporate client onboarding and endorsement, orchestrating document processing, insurer quotation comparison, benefit analysis, recommendation generation and multi-stage workflow execution.
- Developed a LangGraph-based insurance assistant using document-grounded RAG, embeddings, semantic retrieval and tool-driven workflows to answer questions about health insurance quotations and policy benefits.
- Developed components of an agentic workflow for financial claims recovery, automating document review, case analysis, structured decision steps and multi-stage processing with controlled workflow execution.
- Built an enterprise-wide document extraction service using fine-tuned open-source multimodal models to extract text and structured fields from sensitive documents, keeping processing on-premises.
- Developed and optimised GPU-accelerated inference APIs with FastAPI and Docker, reaching sub-5-second inference for image-based extraction workloads.
- Built evaluation, monitoring and observability workflows measuring extraction quality, structured-output consistency, latency, failure patterns and production reliability.
AI/ML Engineer · eData Information
Jun 2023 – Jun 2025
Dubai, UAE
Built and deployed production AI systems for automotive valuation, vehicle data intelligence and insurance automation.
- Developed and maintained a production vehicle price prediction engine using ensemble regression and deep learning, with automated daily market-data ingestion, feature engineering, model evaluation and valuation updates.
- Built VIN and chassis-number recognition systems using computer vision and deep learning to extract vehicle identifiers from images.
- Delivered vehicle valuation and VIN recognition as scalable commercial APIs integrated into the company's automotive and insurance products.
- Designed automated pipelines for ingestion, web scraping, validation, deduplication and feature engineering, turning large volumes of automotive listings into structured datasets.
- Developed RAG applications and agentic workflows, fine-tuned open-source LLMs, and deployed containerised AI services.
AI/ML Engineer & Data Scientist · Independent contractor
May 2021 – Jun 2023
Oran, Algeria · remote
Project-based ML and data engineering for startups and small businesses: predictive analytics, automation and decision support.
- Built ML models for regression, classification, forecasting and decision support.
- Owned data cleaning, feature engineering, model selection and hyperparameter tuning.
- Built ETL pipelines turning raw data into structured datasets for ML workflows.
- Evaluated models with cross-validation, error analysis and task-appropriate metrics.
Full-Stack & Mobile Developer · Independent contractor
Mar 2018 – Apr 2021
Oran, Algeria · remote
End-to-end web and mobile applications for startups and SMEs across the MENA region, with clients in Saudi Arabia and Algeria.
- Built cross-platform mobile apps with Flutter, backend APIs, authentication and databases.
- Developed REST APIs with Django, Node.js and ASP.NET, including third-party integrations.
- Designed PostgreSQL and Firebase data models and ran deployment and maintenance.
What people say
“Atmane is a highly skilled AI and software engineer with a strong ability to turn technical concepts and specifications into robust, production-ready AI products. What particularly impressed me is his ability to cover the full development lifecycle, from understanding the requirements and designing the solution to implementation, integration, and deployment. During my time at First.tech, we worked together on several AI initiatives, including bilingual Arabic-English conversational and voice bots, OCR solutions for Arabic and English, and fraud detection solutions. Atmane played an important role in developing these solutions and consistently demonstrated strong technical expertise, problem-solving skills, and a pragmatic approach to delivering working products.”
“I had the pleasure of working with Atmane and found him to be a highly professional, knowledgeable, and dependable colleague. He has a strong technical understanding, approaches challenges with a positive attitude, and consistently works toward finding practical solutions. Atmane is also a great team player who communicates well and is always willing to support others.”
“Atmane is a versatile AI/ML Engineer with a lot of real-world experience making AI systems that can handle thousands of users and are ready for production. We have worked together on several projects, and I have always been impressed by his knowledge in both AI and software engineering. He is collaborative, thoughtful and has a talent for solving complex problems.”
“We were on the same team building an automated system that involved ETL pipelines, data processing, and machine learning models for forecasting and classification. What I appreciated most about working with him was how easy he was to collaborate with and how willing he was to jump into technical problems and figure things out with the team. He has a strong understanding of AI and machine learning and brings a good mix of AI, data, and software engineering skills.”
Skills
- LLM & agentic AI
- RAGLangGraphMulti-agent systemsTool callingFine-tuning (LoRA, QLoRA)Embeddings & semantic searchMultimodal modelsHugging Face Transformers
- Machine learning
- Regression & classificationEnsemble methodsForecastingComputer visionNLPPyTorchTensorFlowscikit-learn
- Serving & optimisation
- vLLMGPU inference optimisationQuantisationLatency optimisationFastAPI
- Data & backend
- PythonSQLPostgreSQLPandasData pipelines & ETLMicroservicesDjangoNode.js
- MLOps & infrastructure
- MLflowDockerKubernetesCI/CDPrometheusAirflowAWSAzureGCP
Open source
Jev-AR ↗
322M Arabic-first routing model for agents and RAG: MSA, Gulf dialects and code-switched Arabic, ~10 ms per query, runs locally.
evalcascade ↗
Open-source adaptive evaluation for LLM, RAG and agentic systems.
jev-mem ↗
Decision-native long-term memory for AI agents.
canirunthismodel ↗
Paste a Hugging Face model and find out whether your machine can run it locally.
video-gen-automation ↗
Fully local AI video pipeline: topic in, 1080p MP4 out.
word2vec-from-scratch ↗
Word2Vec in PyTorch from first principles: skip-gram, negative sampling and CBOW.
glove-from-scratch ↗
GloVe in PyTorch from first principles: co-occurrence matrix and weighted least squares.
Education
MSc, Artificial Intelligence
University of Science and Technology of Oran – Mohamed Boudiaf (USTO-MB), Algeria · 2020–2022
Thesis: a machine-learning time-series forecasting system for regional rainfall prediction using real-time NASA meteorological data.
BSc, Computer Science
University of Science and Technology of Oran – Mohamed Boudiaf (USTO-MB), Algeria · 2017–2020
Thesis: a recycling platform with web and mobile apps connecting households with waste collectors.
Certifications
- Certified Data Science, Data Analytics & Machine Learning Professional · LIRC, Dubai
- Advanced Learning Algorithms · DeepLearning.AI & Stanford
- Unsupervised Learning, Recommenders, Reinforcement Learning · DeepLearning.AI & Stanford
- Generative AI: Introduction and Applications · IBM
- Introduction to Artificial Intelligence · IBM
Languages
- Arabic · Native
- Berber (Tamazight) · Native
- French · Fluent
- English · Fluent