A About
Closing the gap between a demo and a system you can trust
I build and run the internal systems of a performance-marketing agency, alone. I joined as IT staff; what I actually do now is different — turning work that used to run on human hands (reporting, compliance, lead QA, budget pacing) into systems with tests and gates.
On the personal side, I operate 58 MCP tools and 150+ autonomous loops on a single Mac, for months on end. That experience left me with one conviction: the hard part is not the model — it is the orchestration, governance and cost discipline around it. That is the gap between an impressive demo and a system you can trust, and what I build lives exactly in that gap.
It is also why every number on this site carries ameasuredestimated label. Overclaiming is borrowing against trust.
P Principles
- fail-closed
- Expired compliance IDs, missing required fields, failed checks — whatever a human might miss, the pipeline dies loudly instead.
- governance-as-code
- Approval gates, audit logs and budget ceilings are the ground floor of the architecture, not an afterthought.
- honest numbers
- Every figure is labeled measured or estimated. Values that cannot be computed are never disguised as zero.
- local-first, $0
- Local LLMs plus free tiers fix the cost structure itself — savings by design, not by discipline.
N Now
Currently: pushing local-LLM utilization to its limit (constrained decoding, local rerankers, MLX) and extending the public RAG-eval demo to the cloud (pgvector). Public work lives on GitHub.