> Reflect → Think → Act → Iterate → Verify
VC investor turned founder. Building AI systems that augment human judgment, not replace it.
Thesis: Autonomy hits a ceiling at judgment.
Augmentation is the moat.
I design Claude Code architectures that turn solo founders into small, effective teams — decision frameworks, multi-agent patterns, and operational systems that compound over time.
Domains: Healthcare, compliance, SaaS, deep-tech
| Project | What it does |
|---|---|
klirs |
Self-hostable AML/KYC screening that refuses to silently clear. Playwright captures timestamped evidence per check across sanctions / PEP / registry / adverse-media sources; tri-state classifier surfaces ambiguity for human review; PDF + pre-filled Latvian Bar annexes ready for filing. Hosted at klirs.eu. |
Solo-leveling-the-system |
Solo Leveling-inspired gamification for knowledge workers. Tracks real activity as RPG stats with a dark sci-fi HUD, stat decay, streak multipliers, and a 3D awakening sequence. |
/spec |
Structured specification discovery. Turns vague requirements into buildable scope before you write a line of code. |
1. Truth — first principles, verified facts
2. Velocity — when reversible; otherwise Impact
3. Impact — magnitude × likelihood
4. Empathy — people consequences, dignity preserved
5. Aesthetics — clarity as elegance
How I classify decisions
┌─────────────────────────────────────────────┐
│ Is this reversible? │
│ │
│ YES (Type-R) NO (Type-I) │
│ │ │ │
│ Confidence ≥ 0.6? Full deliberation. │
│ │ │ Evidence. Values. │
│ YES NO Risk. Impact. │
│ │ │ Confidence. Effort. │
│ Proceed. State the │
│ tradeoff, │
│ make the call. │
└─────────────────────────────────────────────┘
The uncomfortable question I ask every project
"What would have to be true for the opposite to make sense?"
If you can't answer that, you don't understand your own position well enough to hold it.