Freelancer profile translated to English.
Back to original languageAbout Darius
Portfolio: dariusgros.dev
Site: ourkat-technologies.fr
Resources and blog: ourkat-technologies.fr/blog
ML/Data Engineer with 5 years of experience, including a strong MLOps component: pipeline industrialization, CI/CD, versioning, reproducibility, and model deployment under constraints (GDPR, sensitive data, sovereign infrastructure). Accustomed to analyzing existing models to identify areas for improvement (performance, reliability, costs) and integrating HuggingFace models into real products via Docker. Solid mastery of Python, large-scale data pipelines (Spark, Delta Lake, Airflow), and software engineering best practices (hexagonal architecture, TDD, SOLID). Autonomous, curious, comfortable in tight-knit technical teams. If alone on a project, I can own the full-stack product (back, ml/dl, and front).
Key Skills
Machine Learning & Deep Learning: Python, PyTorch, scikit-learn, classification, regression, clustering, feature engineering, data augmentation, cross-validation, threshold optimization, loss functions, forecasting, A/B testing.
AI / LLM / NLP: Fine-tuning (CamemBERT, QLoRA), transfer learning, RAG, LangGraph, LangChain, prompt engineering, embeddings, pgvector, ChromaDB, Claude API, Ollama, HuggingFace Transformers, sentiment analysis, text classification.
MLOps & Experimentation: MLflow, experiment tracking, model versioning, feedback loops, config-driven training, threshold sweeps, data leakage prevention, reproducibility, Docker packaging, monitoring.
Data Engineering: Apache Spark, Delta Lake, Databricks, Airflow, PostgreSQL, BigQuery, Redshift, ETL pipelines, data modeling.
Backend & Infrastructure: FastAPI, Pydantic, REST API, WebSocket, hexagonal architecture, TDD, SOLID, Docker, Traefik, MFA/OTP, GDPR compliance.
Cloud & DevOps: AWS, GCP, GitHub Actions, CI/CD, VPS deployment.
Languages: Python, SQL, Scala, R
English
Native or bilingual
French
Native or bilingual
Can work on-site
Lille (up to 10km), Nantes (up to 10km), Paris (up to 10km)
Experience
- Ourkat TechnologiesML / AI Engineer (Freelance)CONSULTING AND AUDITSFebruary 2026 - Today (6 months)Lille, FranceEnd-to-end AI systems for SMEs. 3 ML projects + 2 websites delivered.• Design of a complete ML pipeline: ingestion, document segmentation, embeddings, similarity search with pgvector, evaluation, and generation. LangGraph orchestration in a hexagonal architecture.• Hybrid RAG system (BM25 + pgvector proximity search) for automatic citation of sources in generated reports. Two-layer architecture: report skeleton generated by code + optional AI writing module activated after business validation.• Industrialization with Docker packaging, FastAPI integration, PostgreSQL, deployment on sovereign infrastructure, and CI/CD.• MLOps: observability with Langfuse, QLoRA fine-tuning, optimization of inference costs and API latency.→ RAG pipeline for automatic citation of technical documents. Conditional AI writing with business validation.
- Ourkat TechnologiesCo-founder & Lead ML/Backend — KorusFILM AND AVJanuary 2026 - May 2026 (4 months)Lille, France• Fine-tuning DistilCamemBERT for binary classification of toxic content (recall > 0.95). 6 detection categories, threshold optimized to 0.3, weighted CrossEntropyLoss.• Multi-source data generation pipeline: synthetic data from Ollama, web scraping, adversarial augmentation (leetspeak, unicode confusables, zero-width chars). Train/test/val split before augmentation. Zero data leakage.• Config-driven experimentation framework (YAML + MLflow), model versioning with rollback. Production feedback loop: false negatives reported by operators reintegrated into training.• Real-time backend with FastAPI, WebSocket, and PostgreSQL, deployed on sovereign EU VPS, GDPR compliant, without US cloud dependency.→ Fine-tuned model in production. 3-layer moderation pipeline scaled for 100 to 5,000 participants.
- Client PME confidentielAI Consultant, Document AutomationENERGY AND UTILITIESDecember 2025 - February 2026 (2 months)Paris, France• Deployment of a hybrid parsing + LLM system for automatic conversion of technical PDF reports (geotechnical, multi-vendor) into structured JSON/Excel. Modular prompts of 300+ lines with validation by physical domain constraints.• Token optimization: text-based extraction via pdfplumber (no vision API), page-by-page processing. Conservative rejection strategy prioritizing accuracy.• Complete infrastructure: VPS, Docker, LLM API integration, MFA/OTP authentication, GDPR compliance. Gradio interface adopted daily by business teams.→ Fully automated manual processing. ~30% savings on LLM API costs.
Recommendations
Be the first to recommend Darius
Help this freelancer shine by sharing your experience working together.
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Master MIASHS Machine LearningUniversité de Lille2022Relevant coursework: Time Series Forecasting, Advanced ML Algorithms, Mathematical Optimization, Cloud Computing
- Bachelor In Applied Mathematics & Minor in EconomicsWingate University2020Relevant coursework: Data Analysis, Advanced Statistics, Econometrics, Economic Forecasting Models