Open to opportunities

abdelatif touati

@abdelatiftouati

Message

AI Engineer at DIR'AJ who shipped Eidelia, an on-premise legal AI assistant, and created RAGnarok-AI.

France

Message

At DIR'AJ, I designed and shipped Eidelia, the legal case management platform’s AI assistant, on-premise on a CPU-only server. Its hybrid search combines Qdrant and Elasticsearch, with agents for document analysis, statistics, and assistance.

I raised the golden-set score from 77% to 93%, and the deterministic pre-router selected the right agent 85% of the time, compared with 52% using the LLM alone. I also added prompt-injection guardrails and role-based access control.

I created and maintain RAGnarok-AI, an open-source framework for evaluating RAG pipelines and AI agents. It includes retrieval metrics, LLM-as-a-judge, CI/CD quality gates, and production monitoring, and is published on PyPI.

At Sup de Pub (OMNES Education), I designed and taught a 14-hour data science and machine learning course for Master's students. At Wild Code School, I supported data bootcamp cohorts and built the Iliad data platform using Prefect, dbt, MongoDB, and DuckDB.

Experience

Work history, roles, and key accomplishments

Current

Creator and Maintainer

Current

RAGnarok-AI

Jan 2026 - Present (10 months)

Created and maintained RAGnarok-AI, an open-source RAG evaluation framework published on PyPI. Implemented retrieval metrics, LLM-as-a-judge, CI/CD integration, and Helm chart.

Python RAG LLM Evaluation LLM As A Judge CI CD Helm PyPI Pytest Mypy Ruff Pre Commit Qdrant Pgvector Lang Chain LlamaIndex Ollama OpenAI

AI Engineer (Freelance)

DIR'AJ

Feb 2026 - Sep 2026 (7 months)

Designed and shipped Eidelia, the platform's AI assistant, fully on-premise on a CPU-only server. Led architecture, AI agents, document analysis, evaluation, and quality/security improvements.

fastAPI Symfony Ollama Qwen Qdrant Elasticsearch Redis Celery SSE Python Docker Prometheus Grafana RAG AI Agents LLM Evaluation Prompt Injection RBAC

Data Teaching Assistant

Wild Code School

Jan 2024 - Dec 2025 (1 year 11 months)

Supported Data bootcamp cohorts with live coding, project mentoring, lessons, and mock exams in Python, SQL, and Machine Learning. Built the Iliad data platform using Prefect, dbt, MongoDB, and DuckDB.

Python SQL Machine Learning Prefect DBT MongoDB DuckDB

Data Science & Machine Learning Lecturer

Sup de Pub (OMNES Education)

Oct 2026 - Present (1 month)

Designed and taught a 14-hour course for Master's students in Data Marketing & AI, covering decision trees, model evaluation, and customer segmentation on Google Colab.

Data Science Machine Learning Decision Trees Model Evaluation Customer Segmentation Google Colab