The Showroom — AI Cost Infrastructure · The Reasoning Layer · Consumer Products · Business Tools · Care · Free Resources · Education · Energy & Place
What the studio can sell, what it invented to build it, and who it builds for — organized the way an investor reads: the money layer first, the original methodologies behind it, live consumer products, the business tools, care, free help-first resources, education, and energy and place. Every entry carries one honest status tag — the tag means exactly what it says.
Statuses below match the tags on their cards in the sections that follow.
Live = deployed or in real use. Built = real code exists today. Now onboarding = onboarding hosts and clients now, never claimed as shipped. Concept-design = the design, spec, or brief is the artifact today. Projects marked * keep their source private to protect IP — the card is the showcase, live demos are linked wherever one exists, and code review is available on request. Commercial terms are set per engagement; no price is printed for anything whose terms are not final.
AI spend is a bill every business now pays. This layer is the stack built to shrink it — routing, market pricing, and the public home of token conservation and optimization.
For: any business running AI models at volume
An OpenAI-compatible LLM routing platform: prompt compression, personal-data redaction, cheapest-capable-model routing, budget ceilings, and auditable savings proofs. Every call produces a receipt — the savings claim is a ledger, not a slogan. Charged by the tier, never by the token. The interface demos are live; the hosted API is mid-rebuild.
For: AI buyers, traders, and analysis subscribers
A marketplace and analysis platform for AI inference — inference book, live token pricing, and a proving ledger that records only genuine routed receipts. DAR, its market-intelligence component, runs a FastAPI backend (LLM router, auth, Stripe, Supabase, PDF export) generating daily market-analysis reports from technical, on-chain, and sentiment data; its front end is mid-migration. AIM's public site serves a stale build and is deliberately not linked — this card is its current showcase. Its compute-unit engine also carries the studio's humanitarian mission slot, Care for Jajja.
The public home of token conservation and optimization — the education library and the AI cost-audit offer, with the Skool community as its community arm and the HF Incident Timeline hosted on its .tech domain. Its founder carries the credential line "Certified LinkedIn AI Trainer" (Portuguese first, then Spanish and loss-prevention trainer projects, assessments required for access).
The invention shelf — the named, dated methods the products run on, and the research systems where the next products come from. In this studio the stack is one system: Step Zero decides, Ternary judges, RelAI verifies, Masquerade protects, the Router saves.
Negative Step Zero decides — a 10-check pre-action gate (what / where / when / delivery) before any tool runs. Ternary judges — every claim gets +1 / 0 / −1, and 0 stays 0 until a human decides. Tri-Two/.32 sequences the stack — ternary judgment first, then binary execution. RelAI Race verifies — 2–4 models race the same task (lean 2 / balanced 3 / deep 4) and widen only on informative disagreement; a v0.1 harness exists. Masquerade Party Protocol protects — emails, phones, account numbers, and keys are swapped for substitute tokens before any external handoff, in a client-held-passphrase encrypted vault format (AES-256-GCM, scrypt) built on standard, third-party-verifiable cryptography. Relay Race executes — tasks decompose into small segments separated by reversible gates, each run by a fresh agent. BabyG Core is the parent decision path that runs them in order. Operationalized today in the PreFlight Checklist tooling, in active use.
The governed agent-execution kernel the methodologies run inside: preflight checks, authority tiers, a receipts engine, and continuous evidence. Slices 0–2 plus hardening verified at 59/59 tests on the branch linked below; the main branch is not yet assembled, and the repo says so itself.
A text-to-agent control line: one number routes requests to specialist agents under authority tiers, with an identity allowlist and a full audit row per action. Slices 0–1 are in open pull requests, merge pending.
A private, chaired multi-agent boardroom: agents from different providers check in and report work, which counts only after verification against the artifacts — the founder is final certifier. Agent governance practiced daily, not theorized.
For: teams giving AI agents persistent memory
A lightweight REST API for agent memory — API-key auth that fails closed, rate limiting, validation, structured logging. Finished service code (v1.1.0), readable end to end.
A living evidence registry of recurring failure modes across frontier AI models — observation / evidence / inference / uncertainty, with a mitigation per finding. Active research; the repo is the artifact, and it feeds the Frontier Fault Fridays analysis line.
A self-contained interactive timeline of a major Hugging Face security incident \u2014 sourced, structured, and shipped as a public web product with a documented deploy procedure and an integrity note. Documenting and reporting the details \u2014 a public research record of AI agent incidents.
Proves: forensic incident analysis \u00b7 turning research into a product people read
The studio's agent-evaluation test range: task batches run against agents under a sealed-program standard — the program is sealed first, agents verify blind, results are unsealed and compared, disagreements go to a third party or the founder. Batch 001 is complete and mirrored in Roman_Commons; the home repo is currently ruleset-locked, which is stated here rather than hidden.
The research home: three longstanding open problems — the Perfect Cuboid, the Scholz–Brauer addition-chain conjecture, and Markov–Frobenius uniqueness — worked under Engineering by Subtraction (counterexample-first, +1/0/−1 evidence states). The README claims no solutions, and neither does this page.
The QST monorepo — the theory family behind the Router's fold / subtract approach to token cost. In the studio's own inventory this repo is recorded as not yet re-checked after an early sweep hit a rate limit, so it is flagged Unverified here rather than oversold; the theory's working pieces are the ones demonstrated in the Router and AIM engine cards.
Products built for the public — live, clickable, and understandable in five seconds.
A schema-driven field-data platform: a camera-ingestion app ingests public live-camera feeds against one metadata taxonomy, and the public SkyMap register publishes timestamped, sourced aerial-activity reports correlated against satellites, aircraft, and space weather — explained cases published alongside unresolved ones.
For: anyone reconnecting with friends, classmates, and creators
Type a first and last name — FindYaFavs surges every major platform for their public socials, and nothing is treated as a match until the user confirms it. Launched October 31, 2026: a consumer social-search product with a built-in confirmation gate, searching public accounts and pages only.
An agents-first virtual city block — agents as citizens, customers, and merchants; humans as sponsors and observers. The inhabitable layer over the studio's agent-platform lineage (marketplace → shared agent core → Sub-Verse), designed machine-readable-first. It exists as design records and a teaser cut; there is no repository under this name, and none is claimed.
Proves: multi-agent platform architecture · original product thinking
The most marketable software in the portfolio for small and mid-sized businesses — the stack a property shop, contractor, or service business buys piece by piece: phones answered, sites watched, secrets checked, rent rolls read, compliance kept.
For: property shops, short-term-rental hosts, service businesses
Answers and routes the phone in a voice the client chooses, trained on the client's industry and restricted to words the client approved. Qualifies callers, books appointments, triages guest problems at 2 a.m., escalates emergencies with the facts already gathered, and hands the owner supervised summaries. Now onboarding hosts and clients — not claimed as a shipped product, and no client result is claimed.
For: general contractors and trades businesses
A single-page command center — project list, per-project budget and overview tabs — built and hosted at $0. All demo data is labeled fictional in the README; that honesty is part of the work.
For: the MarciAi stack — on-site awareness
MarciAi's on-site companion ("the eye"): a real monitoring package in two private repos. It has never been run and there is no live link to give; it is sold as a designed component of the stack, at exactly that weight.
For: businesses that need a security control plane
The stack's security control plane — architecture, specs, and invariants, with a beacon baseline and a self-assessment tool still in draft. Presented as one product line; the underlying work lives in three private repos, and it is not a running service.
For: any business checking its exposure after a breach
A complete CLI that checks email addresses against the Have I Been Pwned breach corpus and produces Excel reports. Pairs with the founder's Google Cybersecurity Certificate (2025) as claim-and-artifact.
For: claims, case, and desk professionals
A mobile-first PWA dashboard in daily use by a working claims manager: labeled reminders with quiet hours, an inbox / queue / log desk, and an import format behind a redaction fence that auto-blanks claim, policy, phone, and email data. Her data stays on her own phone by design.
For: property managers and portfolio owners
Ingests RentManager / AppFolio / Buildium exports and detects revenue leakage — vacant units, below-market leases, annualized loss. Written by a property manager, for property managers.
For: New England landlords and property managers
Landlord-tenant and housing compliance law across all six New England states, structured for working managers, with a clear not-legal-advice disclaimer. Maintained as a working reference.
For: any business that hires vendors
A vendor due-diligence and continuous-assurance program design: verify licenses, insurance, and standing before you trust a vendor — and keep monitoring after you hire. Its README calls it a program design / service-launch blueprint and claims no live integrations; neither does this page.
For: service businesses drowning in customer paperwork
One front door for customers to send documents, photos, and job details into an organized business inbox. The artifact today is a detailed build brief — the specification is the work, not a built platform.
For: anyone who works by dictation
"It writes what you meant" — a speaker-intent correction layer for speech-to-text, treating personal vocabulary (names, places, coined terms) as first-class. The requirements document, written from lived dictation experience, is the artifact at this stage.
For: consumers — especially seniors — targeted by text scams
A ternary text filter: unknown-number messages are judged one at a time as they arrive; scams and spam are diverted, threats are preserved as evidence and alerted on, and you always know someone reached out. Full spec, a 35-entry threat registry, and a production-hardening pass exist; per its own README, nothing is built or live yet.
Care — the work aimed at elders, people in recovery, and the people who love them.
A humanitarian mission for elders in Uganda, built into an AI engine rather than bolted onto it.
How it works: Care for Jajja is the first mission slot (FLAGSHIP_1) inside the AI Inference Market's compute-unit engine — 30% of the engine's distribution slots, Genesis-assigned to Uganda, with its own internal split: Elderly Care 40% · Sick & Hospice 30% · Children & Education 30%. The slot is locked into the protocol: it cannot be removed without forking it. The domain careforjajja.com is held for the program. This is a mission encoded in a working system, presented as exactly that.
Proves: mission-driven system design · values encoded as protocol, not marketing
Free apps, programs, and reporting services — built to help people. No charge, ever.
A habit-breaking app: triggers, breakers, and streaks, in a complete signed Android build (v1.0.0) recovered and verified in October 2026. Its launch stalled at the Play Store's beta-tester gate — the founder refused paid phone-farm testers and held out for real feedback. The GitHub repo holds the README and a check-in scaffold; the app itself was never in GitHub. A homage to Michael's mother, who spent his entire life helping him break his bad habits.
Proves: complete consumer app delivery · launch-integrity judgment
A private journaling PWA — complete and installable, verified from the founder's archive in October 2026: offline-first, with its journal stored on the user's own device (IndexedDB + service worker). It has no repository and no hosting yet, which also makes it the shortest path in this portfolio to a shipped product: host the PWA, no app-store gate.
Proves: privacy-first consumer design · complete PWA engineering
A technology-education platform for everyday people — seniors and non-technical learners first: 8 modules behind a persistent English / Spanish / Portuguese toggle; learn when you're ready, not when a course calendar says so. Free for all users, with free consultations for seniors. Completing all 8 modules grants the free version of the studio's online token-education platform and free tech training — the reward for finishing, not a purchase. The program's order of importance, in the founder's words: help people first — money or profit is not, has never been, and will never be first, second, or third. Program designed and a trilingual site preview built; the platform itself has not been launched yet.
A weekly Substack translating AI, robotics, and nanotech news into what it means for wages, rent, bills, trades, and small shops — primary sources, independently verified. Its public repo is the research and tracking base, including the medical-breakthrough watch that feeds Hope_Ai.
The evidence base for Breakthrough Watch: real medical and scientific breakthroughs tracked on an evidence ladder — sourced, graded, no hype. The public repo is the artifact; its open pull requests form one sequential citation chain awaiting a single merge decision.
Proves: evidence discipline · health-information curation
Education — plain language, at the learner's pace, in more than one language: the free public programs, and the training and onboarding portals the studio sells to companies for their own staff.
A technology-education platform for everyday people — seniors and non-technical learners first: 8 modules behind a persistent English / Spanish / Portuguese toggle; learn when you're ready, not when a course calendar says so. Free for all users, with free consultations for seniors. Completing all 8 modules grants the free version of the studio's online token-education platform and free tech training — the reward for finishing, not a purchase. The program's order of importance, in the founder's words: help people first — money or profit is not, has never been, and will never be first, second, or third. Program designed and a trilingual site preview built; the platform itself has not been launched yet.
Corporate training in token conservation, AI optimization, and security — sold to companies for their own staff, delivered as that company's own branded training portal.
A custom new-hire onboarding platform, branded to the client company, on the same curriculum spine as the training portal.
A free Skool community and curriculum teaching everyday people and small businesses how AI token costs actually work — no panic, no jargon. The community is live and running; the repo holds its curriculum and operations.
The trades, explained so a non-tradesperson can actually understand them — a member of the Tech When You're Ready family. It exists as a registered domain (held to May 2027) and a design; there is no repository under this name and the earlier site is offline, stated plainly.
A learning platform for young creators: instead of a fixed curriculum, each creator's persistent virtual world grows from their projects — buildings are projects, progress lights the windows, and the city becomes a living portfolio. Progression follows demonstrated capability, not grade ceilings, and safety / privacy documentation is part of the design from day one.
The studio's systems library — methodologies and dashboard concepts in one catalog. A draft pull request seeds its CATALOG.md and awaits a merge-or-close decision; it is listed as the education shelf's reference spine, at exactly that status.
One initiative, included because its subject is energy infrastructure and regional economics — and because it shows the same method applied to physical assets: reuse what already exists before building anything new.
A public-benefit concept for Maine: coordinated tourism, infrastructure reuse, and local production — "Every weekend in Maine, planned. Every old asset reconsidered." The site is live; the initiative itself is a concept open for review by Maine town officials, state agencies, and economic organizations.
The example that matters here — Mill Site Data Farms: Maine's retired paper mills are, on inspection, close to ideal homes for edge computing: heavy industrial power already in place, river water for cooling, heavy-load floors and high ceilings built for paper machines, and room to expand — in the towns that once ran them (Millinocket, East Millinocket, Lincoln, Old Town, Dexter on the Penobscot Corridor; Rumford, Mexico, and Andover in Western Maine). Landmark buildings maintained instead of decaying, property returned to productive, tax-paying use, technical jobs in mill towns. An open question, stated as one: where a site or its surrounding land does not fit computing, could it support solar fields instead? That question is part of the conversation this initiative wants with energy professionals — it is not an existing claim of the site. The initiative's other showcases are Road Trip Atlas (tourism coordinated across towns instead of towns competing) and NorthSpore Grid (distributed local production fitted into reused barns, warehouses, and mill space).
Proves: infrastructure-reuse thinking · regional economic design
Michael F. Chaves · Founder, Subtract Architect Studios · Concord, NH
Direct line: 617-230-0760 — for pitch deck reference or questions about this portfolio.
Investor Portfolio View: michael-chaves-portfolio.vercel.app/investor · Résumé portfolio: michael-chaves-portfolio.vercel.app