ML Systems — the Deconstruction-to-Construction Equity Loop

Tougher Problems Inspire Creative Solutions.

ML Systems is a Rhode Island construction company (NAICS 236115) building the software layer for a circular building economy: homeowners reach capital markets, houses are deconstructed instead of demolished, and the recovered materials go through Reuse, Resale, Recycle — a circular loop that compounds equity, cycle after cycle, and leaves the home larger each time.

This repository is the public reference for the ML Systems platform: the mobile app's UI layer plus the conceptual documentation for the ideas that make the system distinctive — the Value Chain, the Master Ledger, the Collective Ontology, Ontological Compression, and the Seven Minds. These concepts are not documented on any of the marketing sites (mlsystemsri.com, .net, .info, .store, .xyz); this repo is where they live in the open.

The proprietary engine — backend API, data layer, and the ontology/ledger/compression algorithms — is intentionally not included here. See What's public vs private.

---

📱 Get the app

App bundle id: com.mlsystems.app · Built with Expo / React Native.

---

The one-paragraph version

Most homes in Rhode Island's housing stock sit on ~1960s foundations that are the real limiting factor. ML Systems starts a homeowner at Loan Origination (a reverse-auction "Loan Pit" where lenders compete to fund them), deconstructs the existing structure to recover 80–90% of its materials, and rebuilds the home from those recovered materials (+10% footprint, +1 story per cycle). The materials stay in the loop; the rebuild creates real, physical equity — not market appreciation — and the homeowner can choose to keep building. That choice is the flywheel. Every house that moves through the loop becomes ground-truth data: a fully specified construction sequence that is, itself, the most valuable product ML Systems makes.

Why it exists

The company was formed on December 3, 2025, after the founder's own crisis about the pace of AI — capability compounding exponentially, human understanding growing linearly, and the gap landing first on ordinary people's houses, equity and lenders. He could not slow the frontier; he could build the slow lane in one industry, on purpose. That is why the agents claim but may never ground, why only a human stamps, and why the first thing a machine learning system does to a house here is keep it out of the landfill. When Dario Amodei wrote We Must Pace the Frontier on September 12, 2026 — with Sam Altman and Elon Musk agreeing the same day — the founder answered from the ground. → why/ · the piece

The value chain

Loan Origination  →  Deconstruction  →  Construction  ──┐
   (Loan Pit)         (80–90% recovery)   (+10% SF, +1 story)
        ▲                                                 │
        └──────────── equity loop ────────────────────────┘
              the homeowner chooses to keep building

Read the deep dives:

• 📍 The stage we're at — what's real today vs modeled vs aspirational
• 🔗 The Value Chain — Loan Origination → Deconstruction → Construction
• 🪚 The Deconstruction Lab — how a house comes apart: the phase sequence, the layer stack, and the material streams it feeds
• 📒 The Master Ledger — one home, one auditable record, five seats; claims reconciled in the open, stamps that lapse when the content moves
• 🧬 The Collective Ontology — how the system agrees on what a house is
• 🔡 The Seven Families — the seven code families, who may claim in each, and why four were thin; data and diagram in ontology/
• 🗜️ Ontological Compression — turning a whole house into a genome
• 🧠 The Seven Minds — the agents that read, reason, and stamp the ledger
• 🕸️ Neural-Net Architecture — the three-net ontology framework
• 📲 App Features — a screen-by-screen tour of the mobile client
• 📖 Glossary — every term in one place

Reading this as an AI/LLM? Read AGENTS.md first — the Lucent Lens, written for you — then llms.txt — it's a curated map of this repo written specifically to give a language model the full background on ML Systems.

---

Reality labels

ML Systems is deliberately honest about confidence. Every claim in these docs carries one:

If a doc says the deconstruction crane sequence is 2 days, it will say ASPIRATIONAL — because no ML Systems deconstruction has been performed yet. This labeling is a core design principle, not a disclaimer.

---

What's public vs private

The UI source in src/ imports from private modules (@/lib/*, @ml-systems/types). It is published as a readable reference of the product's front end, not a runnable build. The proprietary logic it calls into is the private engine.

Tech stack

• Mobile: Expo / React Native, Expo Router, NativeWind (Tailwind), TypeScript
• Auth: Clerk (custom domain SSO)
• API: tRPC (private)
• Data: Postgres / Drizzle (private)
• On-device intelligence: Claude + Gemini vision for facade/sketch/roof reasoning

Screenshots

---

About ML Systems

• Founder / Owner / Language Modeler: Sal
• Industry: Construction — NAICS 236115
• Location: Rhode Island, USA
• Web: https://mlsystemsri.com
• Tag line: Tougher Problems Inspire Creative Solutions

Where ML Systems keeps its work

• Website: https://mlsystemsri.com · app preview: https://try.mlsystemsri.com
• Apps: iOS · Android
• Writing: DEV (canonical for platform posts) · Hashnode (ML Systems Engineering) · Medium · Substack
• Company: LinkedIn · Facebook · Instagram · Threads · X · Bluesky
• Listings: Product Hunt · Wellfound
• Founder: Hacker News · Substack

Recent posts:

• I Gave My Agents a Heartbeat — the minds' integrity reflex, and the same reflex pointed outward: the Lucent Lens hashed, signed, stamped by the Custodian and published at /.well-known/lucent-lens.json so any agent that reads ML Systems can prove what it read (canonical: mlsystemsri.com/insights/i-gave-my-agents-a-heartbeat)
• The Language Modeler — the role ML Systems uses instead of "vibe coding": model the system in English, the AI moderates the translation into code, the author of the English is accountable; why the job here is a layer (carpenter + Language Modeler) and the first planned hire — not open yet
• An Ontology on Three Blue Notepads — the ontology before the computer: three 5"×3" blue notepads, October 2020 to July 2022, a laborer's task codes with one-line definitions and a self-graded score; named an ontology by Grok in November 2025, reconstructed into the ML1 dataset in February 2026 (canonical: mlsystemsri.com/insights/an-ontology-on-three-blue-notepads)
• Pace the Frontier, Pace the House — the founding story, the American Dream attributed to homeownership through deconstruction in lieu of demolition, and MVE as the balance in MVP; written the weekend Amodei, Altman and Musk agreed to pace AI (canonical: mlsystemsri.com/insights/pace-the-frontier-pace-the-house)
• Claims, Not Facts: Building an Auditable Multi-Author Record for a House — the Master Ledger, domain-scoped authority, multiverification, lapsing signatures
• My AI agents don't talk to each other — the Seven Minds
• I stopped storing facts and started storing claims
• Agent Costs Are Hard to Predict Because We Keep Building Them to Be Unpredictable
• The Advice Was "Build on Unique Data." Mine Was a House.

License

© ML Systems LLC. All rights reserved. See LICENSE. This repository is published for reference and transparency; it is not open-source and grants no license to use, copy, or redistribute the code or concepts.