AI & ML
GPT-5.4 vs Gemini 3.1 Ultra vs Claude Opus 4.6: Picking the Right Frontier Model
A benchmark-driven comparison of GPT-5.4, Gemini 3.1 Ultra, and Claude Opus 4.6 with practical recommendations by use case
Apr 13, 202612 min read
intermediate
Hélain Zimmermann
Co-Founder & CTO @ Ailog
I ship AI systems: RAG, agents, fine-tuning. Finishing an MSc in Machine Learning at KTH Stockholm alongside an engineering degree at ENSIMAG Grenoble.
3+Years Experience
—8+Projects
—1Research Publication
—2Master's Degrees
—3+Years Experience
—8+Projects
—1Research Publication
—2Master's Degrees
—3+Years Experience
—8+Projects
—1Research Publication
—2Master's Degrees
—3+Years Experience
—8+Projects
—1Research Publication
—2Master's Degrees
—About Me
Shipping code, not slides. Papers only when they exist.
I'm Co-Founder & CTO at Ailog, where we build AI features for small and mid-sized teams: retrieval over messy internal documents, agent workflows that replace brittle scripts, automation around the tools clients already pay for. Before Ailog I researched privacy-preserving NLP at INRIA Grenoble, and I'm finishing an MSc in Machine Learning at KTH Stockholm alongside an engineering degree at ENSIMAG Grenoble.
PythonPyTorchLangChainRAG SystemsNLPFastAPIDockerPostgreSQLNext.jsTypeScript
Selected Projects
A mix of startup work, research, and older engineering projects.
Ailog – Custom AI Agents
Co-founded Ailog in March 2025 to build AI features for small and mid-sized teams. We integrate LLMs into existing stacks: RAG over internal docs, agent workflows, automation around tools clients already pay for.
RAGLLMAutomationData Science
Visit WebsitePrivacy-Preserving NLP
Research at INRIA comparing sensitive information leaks in NLP models across French and English. Analyzed how language structures impact memorization risks in LLMs.
NLPPrivacyResearchAI
View PaperIoT Home-Automation Hub
ESP32-based system controlling HVAC, 3D printers, and cameras via Home-Assistant. Arduino programming, Docker containers, and 3D modeling for custom enclosures.
IoTESP32Docker3D Printing
RAG-as-Service Platform
Full-stack platform enabling businesses to deploy custom RAG solutions with document management, multi-language support, and analytics.
Full-StackSaaSAI
Visit WebsiteAutonomous Firefighting Robots
Java simulation of autonomous firefighting robots with A* pathfinding, event-driven simulation, and graphical interface for terrain visualization and robot coordination.
JavaOOPAlgorithms
Image Seam Carving
Content-aware image resizing using dynamic programming and GPU acceleration with Numba. Achieved 10-100x speedup through vectorized GPU operations.
PythonGPUImage Processing
RISC-V Processor Design
Designed a RISC-V processor in VHDL with custom extensions, interrupt handling, and peripheral integration (LEDs, HDMI). Validated on FPGA hardware.
VHDLFPGAAssembly
JPEG Encoder in C
Implemented a JPEG encoder from scratch: color space conversion (RGB to YCbCr), DCT, quantization, and Huffman entropy coding for image compression.
CCompressionAlgorithms
Latest Articles
Notes from production: RAG, agents, privacy-preserving NLP.
AI & ML
TurboQuant: How Google's 3-Bit KV Cache Compression Cuts LLM Memory by 6x
Google's TurboQuant compresses KV cache to 3 bits with zero accuracy loss using rotation-based quantization, enabling 6x memory savings for LLM inference
Apr 11, 202610 min read
advancedAI Agents
Multi-Agent Systems in Production: Patterns That Actually Work
Proven architectural patterns for deploying multi-agent AI systems that survive contact with real workloads and real users
Apr 10, 202610 min read
advancedRAG Systems
GraphRAG in Practice: When Vector Search Alone Is Not Enough
How combining knowledge graphs with vector retrieval achieves up to 99% search precision and when you should adopt GraphRAG
Apr 9, 202610 min read
advancedRAG Systems
RAG for Code: Building Retrieval Systems Over Codebases
How to adapt RAG techniques for code search with syntax-aware chunking, code embeddings, and AST-based retrieval
Apr 9, 202610 min read
advancedGetting Started
Running Local LLMs with Ollama: A Practical Guide
How to set up, run, and integrate local language models using Ollama for development, testing, and privacy-sensitive workloads
Apr 9, 20269 min read
beginnerResearch
Research on privacy-preserving NLP, done at INRIA Grenoble.
2024 – INRIA
Towards the Anonymization of the Language Modeling
arXiv • INRIA Grenoble
Two training schemes against memorization in fine-tuned language models: a masked objective for BERT-style models, a causal one for GPT-style. Both target direct and indirect identifiers. Evaluated on a medical dataset against several baselines.
View PaperResume
Experience and education.
Experience
Mar 2025 – Present
Co-Founder & CTO
Ailog • Hybrid
Mar 2025 – Present
Technical Consultant – RAG
Nov 2023 – Mar 2025
Technical Director
Nsigma Junior-Enterprise • Grenoble
Jun – Sep 2024
Research Intern
INRIA • Privacy-Preserving NLP
May – Jun 2023
IT Technician Intern
IE-Concept • IoT & Embedded
Education
2025 – 2026
MSc Machine Learning
KTH Royal Institute of Technology • Stockholm
2023 – 2026
Engineering Degree
ENSIMAG • Grenoble INP
2021 – 2023
Preparatory Cycle
CPP • Valence
Get in Touch
AI engineering, research, or client work at Ailog.
Send a Message
Inquiries, collaborations, or just hello
Prefer dev.to, Scholar, or another channel? See all ways to reach me