Pierre Kasparian, GDPR-compliant AI integration consultant
Fourth-year engineering student at UTT (Université de Technologie de Troyes), specialising in computer science and information systems, I have been working as a freelancer since 2024 on AI integration and data engineering projects for French and European SMBs and startups.
After a six-month internship at Pretto (French fintech, mortgage lending) followed by three months of freelance, I have hands-on production LLM experience: RAG pipelines, multi-provider batch inference services, AI automation, and prompt evaluation with Langfuse. I also worked with LiveSession on a multi-tenant RAG chatbot.
My positioning: AI integration with a non-negotiable GDPR compliance constraint. In practice, this means European hosting (OVH, Scaleway), self-hosted open source models (Mistral, LLaMA) for sensitive data by default, and skills transfer at the end of every engagement.
Background
Linear and logistic regression, SVM, k-means clustering, decision trees, dimensionality reduction (PCA). Implemented with scikit-learn and pandas. Capstone project: stock price prediction using regression on historical data.
Full RAG stack (chunking strategies, embedding, retrieval, cross-encoder reranking), QLoRA fine-tuning on Mistral and LLaMA, autonomous agents with LangChain and n8n, open source model deployment via Ollama. Udemy certification obtained in January 2025.
UTT - Computer Science & Information Systems
5-year engineering programme with a data specialisation. Modules: algorithms, relational databases, software architecture, software engineering, networks, security. Chosen 4th-year specialisation: data engineering and applied AI.
Experience
Pretto
Aug 2025 – Feb 2026AI & Data Intern
Built Python-Airflow-dbt ETL pipelines for production. Multi-provider LLM batch inference service (OpenAI, Anthropic, Mistral, Vertex AI) processing 3,000+ inputs per day. 50% inference cost reduction through dynamic model routing based on server load and token volume.
AI & Data Freelance
Prompt auto-improvement pipeline using Langfuse and annotated datasets. LLM platform hardening: removing base64 processing, centralising third-party clients behind a Factory pattern. Slack-native prompt evaluation platform refactoring, reducing production regressions by 80%.
AI & Data Freelancer
2024 – present10+ projects delivered since 2024: multi-tenant RAG chatbots, ML classification on business data (91% precision), AI automation with n8n, SaaS development with Next.js and FastAPI. Data engineering (SQL, dbt, Airflow) as foundation before any AI integration.
See all my worksJunior Conseil UTT
2023 – 2025Technical Director and Project Lead
Led client projects, managed technical teams and coordinated deliverables. Developed a web application for a solar safety association.
Head of Communications & Events
Managed social media, external communications and annual event organisation for the student association.
My values
Every integration is built for compliance from day one. In practice: hosting provider selection is the first constraint, not the last; self-hosted open source models by default for sensitive data; systematic CLOUD Act assessment before integrating any US API. Compliance is not a cost, it is a selling point.
I transfer skills. At the end of an engagement, you should be able to maintain and evolve what we built together. Every deliverable includes technical documentation and a handover session with the team. If the developers do not understand how the system works, the engagement is not finished.
Fast POC, short feedback loop, iterative delivery. No over-engineering. A pipeline that works in production beats a perfect architecture on paper. I use tools suited to the problem, not the most fashionable.