CryptoInvestYL
Expert AI systems
Automated signals and analysis for multi-portfolio management, with the human decision kept where it counts.
Read the case study →AI & agentic systems
AI serving your workflows, not the other way around.
I design governed AI systems that handle real business processes: specialised AI experts, multi-agent workflows where justified, human validation where it matters.
The right level of AI in the right place
Autonomy is not a goal, it is a parameter. The question is not “can we automate?” but “at what level of autonomy does this process actually gain?”
AI assistant
AI assists a person on a precise task. The human steers, the AI accelerates.
Specialised AI expert
A dedicated AI expert for one domain: analysis, document processing, structured extraction.
Intelligent workflow
Several orchestrated steps, with specialisation and quality control at every boundary.
Agentic system
The system acts on its own within a bounded scope, under explicit guardrails.
Five levels of autonomy — you choose, you don’t endure
Every step of autonomy must pay for itself: in created value, not in added risk. The retained level is justified in the design, not afterwards.
Assist
AI helps, the human does.
Analyze
AI analyses and documents, the human interprets.
Recommend
AI recommends, the human decides.
Act with approval
AI acts, the human approves.
Act autonomously
AI acts on its own, within an explicit frame.
Designed and operated in real conditions
Three AI systems in daily use — not demos, not prototypes. Each one illustrates a level of the taxonomy above.
CIYL Engine
Multi-agent workflow
Editorial content production by specialised agents: separated production and verification, human approval before publishing.
Read the case study →Aurora
Agentic engineering
An agentic engineering environment: specialised agents, persistent state, adversarial review. Open source under the MIT license.
View the project (open source) →My position
The business problem first
AI is never the starting point: the process to improve is. The technology is chosen after understanding the problem.
Every level of autonomy must be justified
Each step of autonomy must pay for itself — in created value, not in added risk. Deliberate, not endured.
Human validation at critical points
The system proposes, the human publishes — where errors are expensive. Checkpoints are part of the architecture.
Cost and observability from day one
Tokens, calls, drift: instrumented, not guessed. An AI system without observability is not a system, it is a promise.
A process to automate?
Describe your workflow: I will come back with a first read of the subject and the appropriate level of autonomy.
Discuss an AI workflow