I turn messy, ambiguous business problems into systems that can actually operate.
My background is not a traditional software-engineering or product-management ladder. I built and ran a company, hired and managed people, worked with enterprise customers, raised external capital, and then moved hands-on into AI-native product building.
The recurring pattern in my work is simple:
listen → decompose → model → constrain → automate → measure → improve
AI is leverage. Product judgment stays human.
I am especially interested in systems where probabilistic AI must coexist with deterministic business rules, auditability and real operational consequences.
~10,000 invoices/month in production.
AI-assisted extraction and classification wrapped in deterministic validation, routing, approval workflows and integrations with KSeF, Monday.com and accounting systems.
Principle: the model proposes; the system validates what is allowed to happen next.
~1,500 people handled with almost no manual operator intervention.
A schema-driven onboarding / recruitment engine where process state and next steps are deterministic. The LLM is deliberately bounded to language generation where flexibility is useful.
Principle: 90% determinism, 10% LLM.
An AI-first feedback system designed around a feedback → triage → engineering context → agent → reviewable PR loop.
The goal is not to collect more tickets. It is to shorten the distance between a real user problem and a verified product change while keeping human review at the code-change boundary.
A primitives-first architecture for constrained AI composition:
Probabilistic intent / planning
↓
abstract Plan
↓
schema + dependency + policy validation
↓
deterministic execution
The core idea is to let models interpret and propose while validated primitives, policies and execution engines control critical actions.
A privacy-first, event-sourced AI product for couples built around context, consent, conflicting perspectives and safe AI mediation.
It is an example of a different reliability problem: not financial rules, but human ambiguity, privacy and trust.
I was founder & CEO of Xpress Delivery, a same-day logistics company. I recruited and managed teams, worked with enterprise customers and investors, and helped build the business from the operating side rather than from a software role.
In 2022 I was recognised by BRIEF among the 50 Most Creative People in Business.
That background shapes how I approach technology today: architecture starts from the operational problem and the outcome that should move — not from a model, framework or trend looking for a use case.
I use AI heavily to compress the cost and time of research, prototyping, implementation and verification. That allows me to test ideas unusually quickly.
But I do not assume AI should own every decision.
I actively separate:
- ambiguity that benefits from probabilistic models,
- rules that should stay deterministic,
- actions that require policy or validation,
- decisions that must remain human.
I am not primarily a coder who happens to use AI.
I am a founder/operator and systems thinker who can now materialize product ideas directly because AI has dramatically reduced the implementation barrier.
Open to conversations around AI product leadership, enterprise agentic systems and 0→1 product creation.