Privacy-first AI engineering
AI software that ships — and the numbers to prove it.
We prove our capability the honest way — by shipping our own tested, code-reviewed products. That same delivery engine is what we put to work for you.
- backend on a shipped product
- 143tests · 97% cov
- across 60 frontend tests
- 0a11y violations
- specialist engineering cohort
- 16agents
- confidential data never leaves the machine
- 100%local
Why MAE-I
Not a consultant who talks about AI.
An operator who commands an AI engineering cohort that demonstrably builds and ships. Five things that make that real:
The engine is the evidence
A cohort of 16 specialist AI engineering agents shipped a production app end-to-end — real tests, real coverage, green CI — and a live-deployed business app. We lead with shipped artifacts and hard numbers, not slideware.
Privacy-first by architecture
A self-hosted AI stack means confidential and regulated work is processed without data ever leaving the machine. It is a structural property of how we build, not a promise on a slide.
Governance built in, not bolted on
Every build runs under test-driven development, a threat-model gate, and a mandatory code-review gate — mapped to OWASP ASVS L2, ISO/IEC 42001, and the NIST AI Risk Management Framework.
Outcomes over hype
Shipped products are shown with metrics. Products still in build are labelled as roadmap — never dressed up as live results. The restraint is the point.
Systems that hold without me in the room
This very site is built and maintained by the same agent cohort via git and CI/CD. The work holds up when the founder is not personally present — that is the model.
Services
Outcomes, not deliverables.
Three productized ways to work together. Every engagement runs under the same gated pipeline we use on our own products.
AI-Readiness Assessment
Know exactly where AI pays off — and where it does not.
A structured review of your workflows, data, and risk posture that ends with a prioritized, costed roadmap you can act on with or without us.
- Workflow + data-flow mapping
- Opportunity shortlist with effort/impact scoring
- Privacy & governance risk read
- A costed, sequenced roadmap
Agentic Build Sprint
A real, tested feature or tool in production — not a demo.
A time-boxed sprint where the engineering cohort designs, builds, tests, and code-reviews a working slice under the same gated pipeline used for our own products.
- Architecture + threat model up front
- Test-driven implementation
- Multi-agent code-review gate before merge
- Runbook + handover docs
Privacy-First Automation Build
Automate sensitive work without your data leaving your control.
For regulated and confidentiality-bound teams: automations built on a self-hosted stack so PHI/PII and client documents are processed locally, never sent to third-party clouds.
- Local/self-hosted tooling design
- Data-handling policy + boundaries
- Human-in-the-loop controls where it matters
- Evidence trail for auditors
Process Excellence — Lean Six Sigma
Find the waste and variation your process hides — with evidence, not opinion.
DMAIC-based process improvement led by a trained Lean Six Sigma Black Belt — trained, not certified, and we say it straight. The agent cohort augments the method: statistics are computed deterministically by real stats libraries and only interpreted by AI, so every claim traces back to the data.
- Current-state mapping (SIPOC / value-stream)
- Baseline + variation analysis
- DMAIC facilitation, Define through Control
- Honest scoping: if Six Sigma is overkill for your problem, we'll say so
Custom or enterprise scope? We size it in a scoped proposal after a short discovery call.
Proof
Shipped work, with the receipts.
Real products with real metrics. Client engagements appear as anonymized capability lines — we never expose a client's name or work.
How we build
A governed delivery pipeline.
Every build moves through the same gates — the discipline is what makes the output trustworthy, not just fast.
- Step 1BootstrapContext, constraints, and confidentiality boundary set first.
- Step 2ShapeProblem framed; scope and success criteria agreed.
- Step 3Threat-model gateSTRIDE + AI-specific risks before a line of code.
- Step 4BuildTest-driven implementation by specialist agents.
- Step 5VerifyThe change is driven end-to-end, not just unit-tested.
- Step 6Code-review gateMulti-agent review of the exact diff — blocking.
- Step 7Operator mergeA human merges only when every gate is green.
16 specialist agents
Architect, backend, frontend, devops, security, QA, data, UX, research and more — each with a defined lane.
~160 skills
A curated library of reusable engineering skills the cohort draws on per task.
Standards-mapped
OWASP ASVS L2 · ISO/IEC 42001 · NIST AI RMF · SOC 2-readiness · C4/arc42/ADR.
Hybrid retrieval
A local knowledge service — keyword + vector + rank fusion + cross-encoder rerank — over project corpora, offline.
Verified research
A research agent with mandatory citation-verification and demonstrated prompt-injection refusal.
Evidence trail
Decision logs, RACI, and activity logs on every build — auditable by design.
Trust
Privacy-first by architecture.
For healthcare, finance, and any team under a confidentiality duty: the tooling is built so sensitive data never leaves the machine.
The boundary
Confidential and PHI-adjacent content is processed only by local and self-hosted tools. It is never sent to a third-party cloud, and never pushed to any public repository.
- Self-hosted web search and crawl (Docker, loopback-bound) — no third-party cloud in the loop.
- Offline document-to-text conversion — files are parsed locally.
- A local retrieval store — indexed content stays on the machine.
- An enforced boundary: confidential and PHI-adjacent content never touches a public API or repo.
Roadmap
What we're building next.
Our own products, in active build. Shown as vision — not as live results, returns, or customer counts.
Meeting-Intelligence SaaS
A multi-agent project-management and meeting-intelligence product with an all-Claude, privacy-preserving router that never trains on customer data.
Self-hosted transcription engine
A no-train, self-hosted meeting-transcription engine with first-class English–Tagalog code-switching, designed to replace a paid cloud service.
Lean Six Sigma DMAIC engine
An end-to-end multi-agent DMAIC consultant — Define through Control to closure — where the statistics come from scipy/statsmodels and the LLM only interprets. Built as an academic capstone on a fully synthetic case.
About
MAE-I Technologies
MAE-I Technologies is an independent, founder-led AI engineering studio. Its work is done by a governed cohort of specialist AI agents — an architect, engineers across backend, frontend, and infrastructure, plus security, QA, data, and research — that build production software under test-driven development and a mandatory code-review gate.
It is founded and led by Butch Martirez, who directs the cohort, owns the client relationship, and stands behind every merge. The thesis is simple: the pure one-person generalist is a fragile model. Durable trust comes from systems that hold up when the founder is not in the room — so that is what we build.
The privacy-first stance is not a marketing line. It is why regulated and confidentiality-bound clients can automate sensitive work with us at all: the architecture keeps their data local by construction.
How this site was built
This site is itself a proof point. It was designed, built, and is maintained by the same MAE-I Build agent cohort — architect → UX → frontend → security → devops → code-review — deployed via git and CI/CD. The exact pipeline we run for client work, dogfooded in public.
Contact
Let's scope your first build.
A short, no-obligation call. We'll tell you honestly where AI pays off for you — and where it doesn't.
Book an intro call
Fifteen minutes to understand your goal and whether we're a fit. No slides, no obligation.
Prefer to write first?
Message us on LinkedInWe reply within one business day. Confidential enquiries are welcome — sensitive materials are handled on a privacy-first, self-hosted stack.