Carlos Arleo

Architect · Regenerative Systems Architect · Urban Researcher · Newcastle, UK

I design systems that hold by structure, not by instruction.

My background is architecture — fifteen years designing cities, housing, and public space across Europe, the Middle East, Africa, and the Americas. That training taught me to think in whole systems: how parts relate, where things fail, what a place is actually for. I now apply the same approach to AI systems, governance tools, and civic technology. The thread connecting all of it is the same question: how do you build something that behaves well under pressure, not just when conditions are ideal?

From buildings to systems

Architecture and AI systems design look like different disciplines. In practice, they share the same underlying problem: you are designing something complex that will be used by people who didn't build it, in conditions you can't fully predict. The methodology transfers.

Architecture
Understand the place before touching it. Map what already exists — social, ecological, historical. Build from what's real, not from an abstract brief. Design for the whole system, not just the object.
What transfers
Reading complex systems. Identifying where the real constraint is. Distinguishing a symptoms from a root cause. Designing for failure modes, not just for ideal use. Building things that can be maintained by others.
Systems design
Understand the problem before writing code. Map the failure modes. Encode the rules that must hold as hard constraints — not suggestions, not preferences. Build instruments that behave the same way under pressure as they do in testing.

What I've built

Four instruments across completely unrelated fields — scientific research, AI governance, music, and civic planning. Each was built from scratch, in a domain I had no prior technical experience in. Each one produced the same underlying pattern: when you encode the rules that must hold as hard constraints rather than learned preferences, the system becomes measurably more robust, more auditable, and more trustworthy.

Community Planning Support System (CPSS)
A tool that helps 700 + community groups participating in UK planning decisions on equal footing with professional developers. It takes local knowledge — things a community knows about their place that no consultant can know — combines it with 54 live government data feeds, and produces a statutory objection letter that meets the legal standard a planning officer is trained to evaluate. The problem isn't that communities don't know enough. It's that they can't express what they know in the form the planning system accepts. CPSS closes that gap. Proof of concept complete.
Proof of concept complete. Independently verified letters scoring 88–94/100 on legal compliance across three test cases (Carrington, Lapwing, Mowbray). Partnership with the Community Planning Alliance (800+ UK organisations). National Lottery Fund application in preparation.
Proof of concept
Darwinian Novelty Indicator (DNI v4.0)
A system that measures whether a scientific paper is genuinely new — not just different-sounding, but actually advancing knowledge. It uses five independent reviewers who each read the paper and argue about it; a second round triggers automatically when they disagree too much. The scoring is transparent: you can see exactly why a paper received the score it did.
Shortlisted in the Nesta / UKRI Metascience Challenge. Scored highest on the benchmark's hardest task (41% vs the previous best of 31%). 100,000 papers processed. Ongoing collaboration with SPRU, University of Sussex.
Shortlisted · UKRI
AURA-ECHO
A neurosymbolic music system — built as a test environment for the constitutional AI architecture — that generates music under adversarial conditions. The interesting result wasn't the music. It was the behaviour under attack: 319 sessions of deliberate attempts to break the system's governing rules, zero successes. The system doesn't just refuse bad instructions; it detects when the load on its decision-making is being artificially inflated and responds before a violation can occur.
Produced the empirical validation of the Constitutional Physics framework. Co-authored preprint with Dr. Tomas Veloz (Vrije Universiteit Brussel). Endorsed by Audrey Tang, former Digital Minister of Taiwan. 1,800+ preprint downloads.
Deployed
300+
adversarial sessions
zero system failures
100k
papers scored
DNI metascience
94/100
CPSS letter score
independent legal review
800+
UK community
planning organisations

Current focus — Community Planning Support System

The CPSS is the most concrete application of the methodology to date. It takes a technical architecture developed across AI safety research and applies it to one of the most practically urgent asymmetries in British civic life.

"Community groups know their place better than any consultant. They just can't express what they know in the form the planning system requires."

When a developer applies to build on a flood plain or a peat bog or a community park, they arrive with a professional legal team, environmental surveys, and planning statements that cite the right policy paragraphs. The community group arrives with local knowledge, genuine concern, and a deadline measured in weeks.

CPSS doesn't replace the community's knowledge — it translates it. You tell the system what you know about your place. It cross-references that against 54 government data feeds, checks it against relevant case law, and produces a letter in the statutory language that a planning officer is required to consider. The letter is verifiable: every claim it makes is traceable back to a real data source or a real legal precedent.

The proof of concept is complete. We are now preparing for the next stage: expanding the community partnership network, securing funding, and moving toward a governed, community-owned service.


CPSS roadmap

Stage What it means Status
Proof of concept System built and producing legally-verified output across three real UK planning cases. Constitutional architecture validated. Partnership with Community Planning Alliance established. Complete
Pilot deployment Selected community groups in the CPA network use the system on live cases. We learn what works in practice, collect feedback, and refine the interface. National Lottery Fund application in preparation. In preparation
Governed service CPSS becomes a community-governed tool — not owned by a company, but by the network of organisations that use it. Transparent pricing, hardship waivers, and an open governance structure that reflects who the system is for. Year 2–3
Open architecture The underlying constitutional architecture — the part that ensures the system can only produce claims it can verify — released as open-source infrastructure for other civic tools to build on. Year 3

Background

The AI work didn't come from nowhere. It's the most recent expression of a methodology that developed across fifteen years of practice in architecture, planning, and regenerative development.

2004 – 2013
Architecture practice — Italy, Spain, UAE, Qatar
Trained in Milan, and Barcelona. Worked at OMA (Rem Koolhaas's office) in the Middle East research division — large-scale urban strategy, masterplanning, infrastructure. National Excellence Scholarship, Polytechnic University of Milan.
2016 – 2019
FaulknerBrowns Architects, Newcastle — urban design and research
Led design on mixed-use, sports, residential, and public realm projects across the UK and Ireland. Started PhD at Newcastle University on urban planning and living systems theory.
2020 – present
Co-founder, Atelier Terre — regenerative design, Morocco
Village-scale regeneration focused on local craftsmanship, circular economy, and cultural continuity. This is where the regenerative development methodology became practical rather than theoretical.
2024 – present
Senior Lead Architect, GreenHouse — Africa and Caribbean
Large-scale masterplans in Zambia, Gabon, Zimbabwe, and Jamaica applying Five Capitals frameworks to model socio-ecological flows. Designed the governance protocols that became the precursor to the constitutional AI architecture.
2024 – present
Founder, Localis AI Ltd — constitutional AI research, Newcastle
Independent research lab. Four instruments built across scientific novelty, AI governance, music, and civic planning. Two ARIA Scaling Trust proposals submitted (£900k total). Co-authoring research with Dr. Tomas Veloz (VUB Brussels). Collaboration with SPRU, University of Sussex.

Publications & preprints

Constitutional Physics: Empirical Validation of Aitiopoietic Cognition in Artificial Governance Systems
Arleo, C. & Veloz, T. · Zenodo · March 2026 · Reviewed by Matthew Prager (NYU)
doi.org/10.5281/zenodo.17692900 1,800+ downloads
Constitutional Physics: Empirical Validation of Aitiopoietic Cognition in Artificial Substrates
Arleo, C. · Zenodo · November 2025
https://zenodo.org/records/19616290