Skip to content
Butch MartirezThe Working Record
Metro Manila · remote-first · US/UK/EU/APAC

Process rigor that actually ships.

I'm Butch — a Lean Six Sigma–trained PM/BA who now builds AI products and automations end-to-end with a governed agent cohort. I measure before I build, I ship in the open, and this page is itself the demo.

7+years

delivery, BA & process roles

4industries

finance · healthcare · gov · enterprise

264tests green

on MAE-I Assist, my studio's flagship build

0a11y issues

WCAG 2.2 AA on this page

01 / How I operate

How I operate

  1. /01

    Measure before you build.

    Baseline first, opinions second. If the data says the problem isn't worth solving, I'll say so — even when that costs me the engagement.

  2. /02

    Ship in the open.

    Test counts, coverage, decision logs, honest roadmap labels. Work you can inspect beats work you have to take on faith.

  3. /03

    Own the outcome.

    Not the hours, not the deck — the working thing and what it changed. Systems that hold up when I'm not in the room.

02 / What I do

Two ways to work with me.

Consulting — via my studio

Engagements through MAE-I Technologies

Pick the one that sounds like your problem — each books a 30-minute discovery call with the topic already filled in.

Delivered under a gated pipeline with an AI engineering cohort — my studio, MAE-I Technologies.

Select roles — FTE or contract

Where I add value inside a team

  • Delivery / project management for digital & AI programs
  • Business analysis: requirements, BPMN, UAT, data migration
  • Process improvement with honest Six Sigma discipline
  • AI-adjacent transformation — the PM who can also build

Open to the right team and problem — résumé shared on request via the contact section.

03 / Proof

Automation — from system to shipped.

Two ends of one craft: the sophisticated agentic pipelines I design for my own products, and the focused automations that run today. Real n8n throughout, on synthetic data — client work stays confidential, always. Each carries an honest, Six-Sigma-style benefit estimate: projected, with the assumptions stated.

Flagship · agentic meeting → deliverable pipeline

MAE-I Assist — a governed agent cohort that turns meetings into PMO deliverables.

The automation architecture behind my flagship product: two lanes (project-brief intake and transcript intake), a cohort of specialist agents — PM, BA, quality, meeting-capture, coordinator, formatter — and eight human-approval gates. A transcript becomes structured minutes, an impact assessment (which documents are affected, decisions needed), and formatted deliverables. Nothing auto-publishes.

  • 2 lanes · 8 human gates
  • PM · BA · MoM · Coordinator · Formatter
  • all-Claude · never trains on your data
A large two-lane n8n workflow — Lane A project-brief intake and Lane B transcript intake — each fanning out to a cohort of agents then merging through a coordinator and a formatter, with eight highlighted human-approval gates.
The architecture — two lanes, an agent cohort, eight approval gates
An impact-assessment deliverable: an amber verdict, an affected-documents-by-owner table, a surfaced conflict, and the decisions the owner must make. All names and documents fictitious.
What it produces — an impact assessment, human-gated
Estimated impact

~40–80 min saved per meeting on minutes + impact tracking — plus a consistent record, an audit trail, and human approval gates. Projected from stated assumptions; to be measured.

solution architecture · my flagship build, rendered on synthetic data — no client, no customer claims

Flagship · private meeting-intelligence pipeline

MAE-I Recall — your phone recording becomes a searchable, minuted meeting, privately.

A recording enters and the pipeline runs entirely on-device: transcription (faster-whisper large-v3), speaker diarization (pyannote), word-level merge, then a project-RAG synthesis into minutes, decisions, and tracked actions — handed off to the Assist lane. Audio never leaves the device. This is genuinely Recall's designed pipeline; the data shown is synthetic and it's still in build.

  • record → transcribe → diarize → merge → synthesize
  • on-device · audio never leaves
  • whisperx + pyannote
A three-lane n8n workflow — an intake control-plane lane, an on-device engine lane (transcribe, diarize, merge, index) labelled audio-never-leaves, and a deliverable-synthesis lane that fans out to a minutes author, impact assessors, and a tracker before a human-gated notify.
The pipeline — intake, on-device engine, deliverable synthesis
A meeting-intelligence output: a speaker-attributed transcript beside auto-generated minutes with decisions, action items, and a risk, under an audio-never-left-this-device badge. All names and content fictitious.
What it produces — an attributed transcript + auto minutes
Estimated impact

~1–2 hrs saved per recorded meeting on transcription + minutes — plus a private, searchable archive; audio never leaves your device. Projected from stated assumptions; to be measured. Recall is in build.

capability demo · Recall's real pipeline architecture on synthetic data · in build — no accuracy or deployment claims

And focused automations that ship today

Same discipline, smaller scope — real n8n workflows, unit-tested, on synthetic data. Any of these transfers to your stack.

Lead generation

Find, qualify & log leads.

firecrawl scrapes a source, Claude scores each lead against an ICP, and the qualified ones write to a Sheet. Any source, any industry.

  • firecrawl → Claude → Sheets
  • 25 tests
An n8n workflow: manual trigger, firecrawl scrape, Claude enrich-and-score, Google Sheets append.

~3–8 hrs saved / 100 leads · consistent scoring, no missed sources (projected)

Content / marketing-ops

One input, many channel drafts.

One long input becomes a LinkedIn post, an X thread, and an email in one voice — each a Gmail draft. Nothing auto-sends.

  • one input → 3 channels
  • 27 tests
  • drafts only
An n8n workflow: a form trigger into a Claude generate-drafts node, fanning out to a Gmail create-draft node and a Google Sheets log node.

~25–55 min saved / input · one voice, human sends (projected)

CRM / pipeline

Move a record — the rest builds itself.

An inbound stage change auto-creates the checklist, populates deadlines, and drafts the notification on a live board. Shown on a real-estate pipeline; in essence any CRM.

  • inbound → AI triage → live board
  • 27 tests
An n8n workflow named CRM Pipeline Sync Demo: a form trigger and a webhook trigger feed a single sync-engine code node.

~0.75–2.5 hrs saved / deal · no double entry, nothing missed (projected)

Research → action

Calls & research into tracked work.

A transcript or brief goes to Claude, which extracts action items — owner, due, priority — plus decisions, then files them to Notion or a Sheet.

  • transcript → tracked actions
  • 24 tests
  • Notion / Sheets
An n8n workflow: a form trigger and an optional firecrawl research node feed a Claude extract-action-items node, fanning out to a Notion create-page node and a Google Sheets append node.

~15–35 min saved / input · owners + due dates, nothing missed (projected)

Also — a full product beyond the flows

The same discipline on a larger, deployed build.

Product build · deployed

Attendance & payroll, ₱0 infrastructure

A deployed attendance and payroll app for a small F&B team — country-correct overtime, remote clock-in, capability-based access. Synthetic demo data. It's now being productized into a hosted multi-tenant SaaS (in build).

Interface preview: a phone clock-in screen for a fictional employee showing a confirmed 8:00 AM clock-in with GPS logged. Demo data only.

  • 94 tests green
  • country-correct OT
  • live deployment
Live demo · you're reading it

This page

Every section runs one modern front-end technique — scroll-driven animation, container queries, view transitions and more — each with an accessible fallback.

See how it works ↓

Also built: a BIR tax record-keeping & estimator kit for PH freelancers (59 tests green · my own IP · not tax advice).

Built with the tools you already use

  • n8n
  • Claude
  • Google Sheets
  • Notion
  • Python
  • OpenAI Whisper

…plus firecrawl, Postgres, pyannote & Meilisearch under the hood.

04 / How I work

One method, any problem.

Step 1

Define

The problem in the client's words, the boundary conditions, and what "done" means — agreed before anything else.

Step 2

Measure

Baseline the current state with real data. No baseline, no claims later.

Step 3

Build

Test-driven, in small verified slices — by hand or with the agent cohort, always under review gates.

Step 4

Verify

Drive the real flow end-to-end. Tests passing isn't the finish line; the working thing is.

Step 5

Operationalize

Runbooks, handover, and honest measurement of what changed — 30, 60, 90 days after.

05 / Where I've been

The arc, without the logos.

Titles, industries, and outcomes — employers and clients stay private by design.

Phase 1 — PM/BA foundation

Business Coach / Consultant

FMCG advisory · 2019–2020

Business simplification coaching for FMCG operators.

Business Analyst

Global investment services · 2020–2022

Requirements, process documentation, and UAT for fund-services operations.

Phase 2 — Six Sigma discipline

Lean Six Sigma Black Belt Consultant (freelance)

Training & consulting · 2021

DMAIC projects and Green Belt coaching after completing Black Belt training (trained, not certified).

Regional Program Manager → General Services Manager

Multinational facilities services, APAC · 2022–2024

Program-managed services for a multinational FMCG account across APAC — ₱50M budget scope, 93% KPI attainment across 27 sites, ₱1.5M in verified savings.

Phase 3 — Digital & process engineering

Senior Business Process Engineer

Fintech / govtech · 2025

BPMN process architecture, BDD/TDD requirements, and eKYC / e-verification proof-of-concepts.

Project Manager, Process & Delivery

Healthcare / HMO benefits automation · Dec 2025 – Jul 2026

Delivery leadership in the health-benefits automation domain.

Phase 4 — Building & consulting

Independent AI Consultant & Builder — Founder, MAE-I Technologies

AI products & consulting · 2025–present

Shipping my own products with a governed AI engineering cohort; consulting on AI readiness, agentic builds, and process excellence.

06 / Credentials

Training & credentials.

Lean Six Sigma Yellow Belt

SixSigmaPH · 2020

certified

Lean Six Sigma Green Belt

SixSigmaPH · 2020

training completed

Lean Six Sigma Black Belt

SixSigmaPH · 2020 — practicing the methodology

training completed

Scrum Fundamentals Certified (SFC)

VMEdu · 2025

certified

BS Service Engineering & Management

Mapúa University · 2014–2018

degree

Agentic AI Accelerator

Xaltius · 2026 — multi-agent capstone on LangGraph

in progress

07 / About

The person behind the record.

I started in operations and business analysis, learned to measure things properly through Lean Six Sigma, and spent seven years delivering projects across finance, healthcare, government, and multinational enterprise. Somewhere along the way I stopped being satisfied with recommending things and started building them.

Today I run MAE-I Technologies — an independent AI engineering studio where a governed cohort of AI agents builds production software under test-driven development and mandatory code review. The discipline I learned measuring factory-floor processes is the same discipline that gates every merge.

I try to keep it simple: metrics over adjectives, roadmap labels over promises. If that sounds like how you want your project run, we'll get along.

Based
Metro Manila, Philippines
Works
Remote-first · US/UK/EU/APAC hours
Languages
English · Filipino
Studio
MAE-I Technologies (founder)
Elsewhere
linkedin.com/in/butchmartirez

08 / How this page works

Every section demonstrates one technique. This is the index.

View Transitions API
The theme toggle morphs light↔dark via document.startViewTransition where supported; instant swap everywhere else.
CSS scroll-driven animations
The top progress bar runs on animation-timeline: scroll(); the career timeline and the method cards enter on view() — pure CSS, @supports-gated, gated so unsupported browsers simply render the final state.
Container queries
The two-track cards in "What I do" reflow based on their container's width, not the viewport.
:has() parent selection
Credential rows highlight when the status chip inside them is hovered — no JavaScript.
Sticky card-deck scene
The method steps stack like a card deck as you scroll — pure position: sticky, the scrollbar is the camera. No scripting, nothing to trap your scroll.
@property typed custom property
The contact button's gradient border is painted from a registered @property angle — a custom property the browser knows is an <angle>, so the conic gradient renders crisply. Static by design.
text-wrap: balance / pretty
Headlines balance and paragraphs avoid orphans — the quiet technique proving craft isn't only animation.
Performance & access as features
Zero-JS baseline (all content renders with scripts off), one-shot non-looping motion, full prefers-reduced-motion paths. Lighthouse and axe runs gate every release — measured, not claimed.

Built and maintained with the MAE-I Build agent cohort — the same gated pipeline I run for client work.

09 / Contact

Let's build something measurable.

A short, no-obligation call — consulting scope or a role conversation, your pick.

Or just tell me what's slowing you down — no tech knowledge needed. I read every one.

Client work is handled through my studio, MAE-I Technologies. Confidential enquiries welcome — sensitive materials are handled on a privacy-first, self-hosted stack.