I work at the intersection of applied AI research, language systems, and experimental software.

My main interests are:

• how LLMs represent and manipulate structure,
• how constrained languages and symbolic systems interact with models,
• how to turn vibe coding into a repeatable engineering method,
• and how to build useful systems out of tools that are still, in many cases, glorified probability engines.

Based in Nantes (EU). Operating on Paris time.

I am particularly interested in constructed and compressed languages as interfaces for intelligence:

• Toki Pona
• Lojban
• pictographic and symbolic systems
• low-bandwidth meaning representation for humans and models

I explore:

• prompt structure
• memory and externalized reasoning
• agent workflows
• model behavior under constraints
• LLM-first discoverability and machine-readable interfaces
• practical limits of “understanding” in code and knowledge tasks

I am interested in treating vibe coding as something more serious than improvisation:

• rapid iteration
• scaffolding and reusable harnesses
• memory and feedback loops
• system design under model unreliability
• building fast without surrendering all dignity

Exploring whether simplified or formally structured languages can serve as better control surfaces for LLMs than ordinary natural language.

Experiments around machine-readable visibility: llms.txt, retrieval-facing structure, model-oriented content organization, and interface design for AI-mediated discovery.

Ongoing interest in whether code agents actually form usable internal models of architecture, constraints, and dependency structure — or merely imitate competence locally.

Using AI tools not just for output generation, but for probing questions about reasoning, compression, abstraction, and system design.

• Website: https://abvx.xyz
• Medium: https://medium.com/@abvcreative
• Substack: https://abvx.substack.com

• LinkedIn: https://www.linkedin.com/in/abvcreative/