Pier-Jean MALANDRINO

CORE

CTO / AI Ambassador for the French Government, OSS maintainer at SCUB

Bordeaux, FR

Joined Jul 2023

https://pjmalandrino.github.io/

About

I am the Chief Technology Officer of a French digital services company, where I drive technology strategy, solution design, and R&D. I am also AI Ambassador for the French government's "Osez l'IA" (Dare AI) plan. My current engineering focus is low-bit LLM quantization. I built LLVQ, an independent from-scratch Rust implementation of Leech Lattice Vector Quantization, including a fused multi-shell CUDA decoding kernel and VRAM layouts for 2-bit weights. The work is published as a preprint (arXiv:2609.02652), with an open model on Hugging Face (Pier-Jean/Qwen3-4B-LLVQ-2bit) and an open-source repository (github.com/pjmalandrino/llvq). It also led to a merged upstream contribution to Hugging Face candle. I am the creator of Docling Studio (github.com/scub-france/Docling-Studio), an open-source visual inspection layer for document parsing, and I advise Karate Labs on UI product strategy and AI.

Stats

Reputation:	2475
Pageviews:	207.3K
Articles:	32
Comments:	6

Expertise

AI/ML

Articles
Comments

Articles

Your Quantized LLM Is Not Slow Because of the Quantization

My 2-bit model was slow because of a 778 MB memory copy per token, not the quantization. Profile what you did not compress.

September 1, 2026

· 2,552 Views

Production Checklist for Tool-Using AI Agents in Enterprise Apps

Tool-using AI agents are production software. The same ops discipline applies, but some traditional assumptions quietly break.

May 7, 2026

· 3,655 Views · 1 Like

Designing Docling Studio: Key Architecture Decisions

Architecture decisions behind Docling Studio: a visual inspection layer for Docling. From dual-engine design to Docker packaging.

April 9, 2026

· 3,586 Views · 1 Like

AI-Powered Triathlon Coaching: Building a Modern Training Assistant With Claude and Garmin

Learn how to use MCP technology in conjunction with a Claude desktop application to create your own personal coach with access to your live metrics.

September 25, 2025

· 7,419 Views · 1 Like

MCP and The Spin-Off CoT Pattern: How AI Agents Really Use Tools

I’ve read many articles explaining what MCP is, but none explore how AI actually handles these server interactions under the hood.

July 2, 2025

· 1,907 Views · 3 Likes

From OCR Bottlenecks to Structured Understanding

OCR errors cascade through RAG pipelines, killing performance. SmolDocling (256M params) processes docs holistically → structured output → better RAG.

June 18, 2025

· 2,719 Views · 3 Likes

The Delegated Chain of Thought Architecture

The D-CoT architecture decouples reasoning from execution in LLM by centralizing reasoning in a "modulith" and delegating execution tasks to specialized modules.

February 18, 2025

· 4,852 Views · 4 Likes

Chain-of-Thought Prompting: A Comprehensive Analysis of Reasoning Techniques in Large Language Models

Chain-of-thought (CoT) prompting enables LLMs to improve their reasoning capabilities. This paper explores various CoT techniques and their practical limitations.

January 20, 2025

· 6,328 Views · 3 Likes

RAG Systems: A Brand New Architecture Tool

Discover how RAG systems boost AI by combining real-time data retrieval with language models for accurate, relevant responses.

October 23, 2024

· 5,946 Views · 5 Likes

Data Pipeline vs. ETL Pipeline

Understand the key differences, use cases, and best practices for real-time data processing and batch data integration.

August 15, 2024

· 7,749 Views · 4 Likes

Data Lake vs. Data Warehouse

Data lakes offer flexibility with raw data; data warehouses provide structured data for quick insights. Each has its own benefits and trade-offs.

April 2, 2024

· 1,603 Views · 3 Likes

OWASP Top 10 Explained: SQL Injection

This post discusses SQL Injection (SQLi), its types, examples of breaches, and prevention methods to protect against this cybersecurity threat.

March 12, 2024

· 15,260 Views · 7 Likes

Architecture: Software Cost Estimation

As technology changes more rapidly, software cost estimation becomes much more important, especially for legacy systems. Discover a new estimation method.

February 27, 2024

· 7,235 Views · 2 Likes

Architecture Style: Modulith (vs. Microservices)

Use Modulith architecture to split your monolith into manageable pieces while maintaining their cohesion and unlocking your app's potential.

January 29, 2024

· 22,281 Views · 5 Likes

AI/Machine Learning Techniques: The Cheat Sheet

This paper is a summary that breaks down the big ideas of machine learning into easy-to-understand parts. Think of it as a quick reference or a cheat sheet.

January 24, 2024

· 2,039 Views · 2 Likes

Navigating the CAP Theorem: In Search of the Perfect Database

This article debunks the myth of a perfect database, explaining that no type can fully achieve Consistency, Availability, and Partition Tolerance at once. "

January 9, 2024

· 4,588 Views · 3 Likes

C4 Model Perspective: The Different Types of Software Architects

The article discusses different types of software architects and their roles in system development, all explained from the perspective of the C4 model.

January 8, 2024

· 6,109 Views · 5 Likes

Architecture Patterns: Sharding

Sharding is a database architecture pattern that involves dividing a large database into smaller, manageable parts called shards to improve characteristics.

January 2, 2024

· 4,067 Views · 2 Likes

Architecture Patterns: Publish/Subscribe

The Pub/Sub pattern enables decoupled, scalable message exchange in distributed systems, but it comes with trade-offs: discover which ones here.

January 2, 2024

· 4,723 Views · 3 Likes

Architecture Patterns : Data-Driven Testing

DDT (Data Driven Testing) enhances testing efficiency by automating with external data, offering broad coverage while needing initial setup.

December 16, 2023

· 5,164 Views · 4 Likes

Evolution of Software Architecture: From Monoliths to Microservices and Beyond

Tracking software architecture evolution from monolithic systems to serverless designs, addressing changing business and technology needs.

November 28, 2023

· 9,271 Views · 7 Likes

Architecture Method: C4 Model

The C4 Model provides a detailed framework for understanding and communicating software architecture, balancing complexity with clarity.

November 23, 2023

· 4,744 Views · 3 Likes

Architecture Approach: Domain-Driven Design (DDD)

Domain-Driven Design: A deep dive into aligning complex software with business needs through strategic modeling and continuous refinement.

November 16, 2023

· 3,843 Views · 2 Likes

Architecture Patterns: API Gateway

An API Gateway optimizes client-service interactions and enhances security, but needs careful design to avoid potential pitfalls. Proper use ensures scalability.

November 15, 2023

· 5,263 Views · 4 Likes

Architecture Patterns: Strangler Pattern

The Strangler Pattern enables gradual, low-risk software migration but demands careful balancing of two systems during the upgrade process.

November 15, 2023

· 3,906 Views · 2 Likes

Architecture Patterns: The Circuit-Breaker

In every architectural design lies a shadowy pitfall. Delve into the lurking dangers within popular patterns, revealing the haunting consequences of misapplication.

November 3, 2023

· 5,313 Views · 2 Likes

Architecture Anti-Patterns: The DARK Side of the Architect

Shed light on the specters that haunt our architectural decisions and understand how to exorcise the demons that threaten our digital sanctuaries.

October 31, 2023

· 4,429 Views · 3 Likes

Command Query Responsibility Segregation (CQRS)

In this article, dive into distinct command/query roles, potential database setups, and the implications of data synchronization.

October 30, 2023

· 4,780 Views · 4 Likes

The Outbox Pattern

Learn how the outbox pattern ensures reliable message delivery across distributed systems, safeguarding and forwarding diverse data types effectively.

October 25, 2023

· 4,560 Views · 2 Likes

Selenium Versus Karate: A Concrete Comparative Approach

This article delves into automated testing using Selenium (with Cucumber) and Karate, assessing their features, tech details, and community support.

October 6, 2023

· 4,874 Views · 3 Likes

Comments

Production Checklist for Tool-Using AI Agents in Enterprise Apps

May 18, 2026 · Pier-Jean MALANDRINO

Thank you!

Rethinking Domain Driven Design: Beyond Tactical Patterns

Feb 27, 2026 · Alexsandro Souza

"It’s not pattern-driven design; it’s domain-driven design."
The truth has been said !

So many poeple are taking everything by-the-book wich leads to over-engineering.

Thanks for the paper, great and serious work.

AI-Powered Triathlon Coaching: Building a Modern Training Assistant With Claude and Garmin

Oct 07, 2025 · Pier-Jean MALANDRINO

Thanks a lot !

I think many things can be built upon this idea :)

Chain-of-Thought Prompting: A Comprehensive Analysis of Reasoning Techniques in Large Language Models

Jan 27, 2025 · Pier-Jean MALANDRINO

LLM "thinking" is indeed a fascinating debate! While CoT is a step forward in structured reasoning, it's just one piece of the puzzle. Memory and context are other crucial aspects - really excited about what Google is doing with Project Titan in this space!

RAG Systems: A Brand New Architecture Tool

Nov 04, 2024 · Pier-Jean MALANDRINO

Totally agree, it is a key concept in LLM field !

Architecture Decision Records

Oct 30, 2023 · Santhosh Kumar Ramabadran

Very interesting, good job !