Adrian RepomantaAI EngineerWork With Me

AI Engineer | AI Systems Builder

Building AI Systemsfor the Future.

I design and deploy AI-powered automation systems, intelligent workflows, and scalable digital infrastructure that help businesses eliminate repetitive work and operate smarter.

View My SystemsWork With Me
  • SYSTEM STATUSBUILDING THE FUTURE
  • AUTOMATIONS DEPLOYEDACTIVE
  • LOCATIONPHILIPPINES → GLOBAL

About

Meet the Builder Behind the Systems

IamaFilipinoAIengineerwithabackgroundincomputerscience,workingacrossGuamandAustralia.Ibuildthesystemsbusinessesrunon—theworkflows,agentsandintegrationsthatquietlydotheworknobodyshouldbedoingbyhand.

It started with curiosity about how digital businesses actually operate, moved through marketing and e-commerce, and landed where the leverage is: automation and applied AI. Every project since has been the same question — what is still manual here, and why?

I build for the boring case, not the demo. Modular workflows, validation before anything downstream trusts the data, fallback logic, and documentation good enough that the system outlives my involvement.

Operating principle

Reliability over complexity. If a system cannot be maintained by the people who inherit it, it is not finished.

  1. Started exploring technology and digital business — teaching myself how online operations actually work.

  2. Entered digital marketing and got my first real taste of automation: connecting tools so campaigns ran themselves.

  3. Built e-commerce and digital projects end to end, learning what breaks when a business tries to scale on manual process.

  4. Went deep on AI and automation systems — LLMs, agents, structured extraction, and orchestration platforms.

  5. Went into production work: automation workflows and AI assistants running for real operations teams, starting in roadside recovery.

  6. Automation roles across three companies in Guam and Australia, moving from automation specialist to AI engineer, while building Repomanta — my own lab for AI-powered business systems.

Technology Stack

The tools behind the systems

Grouped by the job it does rather than by how impressive the row looks. Hover any technology to see what it is actually doing in a build — and if that reads as a capability rather than a product, it carries initials instead of a logo, because it is one.

// category

The reasoning layer — where judgement and unstructured data get handled.

Featured Systems

Systems built, deployed and running

177 workflows in the library, across n8n and Make. Six of them are broken down in full below — select one to see how it works, then browse the whole catalogue underneath.

SYSTEM 01

AI Virtual Office Assistant

DEPLOYED

An AI-powered assistant integrated with Gmail, Calendar and OpenAI to manage business operations and automate repetitive administrative tasks.

Workflow

Problem
Admin work was scattered across an inbox and a calendar. Scheduling, follow-ups and status updates consumed hours a day and nothing was tracked in one place.
Solution
An agent with scoped tools: it reads the inbox, understands intent, books and reschedules calendar events, drafts replies, and logs every action. Anything outbound waits for a human approval step.
Business impact
Administrative work becomes review-and-approve instead of read-and-retype. Nothing falls through the gaps because every request is captured as a tracked action.

The library

177 workflows, built and exported

The full catalogue from my workflow library — read straight from the n8n exports and Make blueprints, not a curated highlight reel. 136 of them open as a node diagram.

n8n
136
Make
41
Integrations
43
  • n8n7 steps

    • Gmail
    • Google Sheets
    • OpenAI
    • Slack
    • +1
  • n8n27 steps

    • AI Agent
    • Gmail
    • Google Sheets
    • HTTP API
    • +2
  • n8n12 steps

    • Airtable
    • Google Sheets
    • HTTP API
    • OpenAI
    • +2
  • n8n7 steps

    • Google Sheets
    • HTTP API
    • OpenAI
    • Slack
    • +1
  • n8n9 steps

    • AI Agent
    • Gmail
    • Slack
  • n8n11 steps

    • AI Agent
    • Google Sheets
    • Notion
    • Slack
    • +1
  • n8n13 steps

    • AI Agent
    • Gmail
    • Google Calendar
    • Notion
    • +2
  • n8n12 steps

    • AI Agent
    • Agent Memory
    • Google Drive
    • LLM
    • +1
  • n8n26 steps

    • Gmail
    • Google Calendar
    • Google Sheets
    • HTTP API
    • +2
  • n8n12 steps

    • AI Agent
    • Gmail
    • Google Docs
    • HTTP API
    • +2
  • n8n8 steps

    • Gmail
    • Google Sheets
    • Notion
    • OpenAI
    • +1
  • n8n14 steps

    • Gmail
    • Google Sheets
    • HTTP API
    • HubSpot
    • +2
  • n8n10 steps

    • Google Sheets
    • HubSpot
    • OpenAI
    • Slack
    • +1
  • n8n7 steps

    • Google Sheets
    • OpenAI
    • Slack
    • Webhook
  • n8n14 steps

    • Gmail
    • Google Sheets
    • Notion
    • OpenAI
    • +2
  • n8n10 steps

    • Gmail
    • Google Sheets
    • OpenAI
    • Scheduler
    • +1
  • n8n13 steps

    • Gmail
    • Google Drive
    • Google Sheets
    • HTTP API
    • +2
  • n8n15 steps

    • AI Agent
    • Google Sheets
    • HTTP API
    • Mailchimp
    • +2
  • n8n16 steps

    • AI Agent
    • Gmail
    • Google Sheets
    • HubSpot
    • +2
  • n8n16 steps

    • AI Agent
    • Gmail
    • Google Sheets
    • HubSpot
    • +2
  • n8n16 steps

    • OpenAI
    • Slack
    • Telegram
    • Vector Store
  • n8n46 steps

    • AI Agent
    • Agent Memory
    • Airtable
    • Gmail
    • +2
  • n8n12 steps

    • AI Agent
    • HTTP API
    • LLM
    • Perplexity
    • +2
  • n8n11 steps

    • HTTP API
    • Notion
    • OpenAI
    • Slack
    • +2

Client-owned blueprints are listed by what they do; names, brands and people are stripped at build time, and their node structure is not published at all — that structure is the client's business logic.

Projects

Things I have built and shipped

Client sites, internal tools, research and hackathon builds. Every screenshot below is the real application running — not a mockup — and every stack list says what the technology is actually doing.

  • Repomanta Digital Lab running locallyLIVE

    Repomanta Digital Lab

    Automation studio — marketing site

    The studio site: services, solutions, case studies and insights, with a prerender step so crawlers and AI answer engines get real HTML instead of an empty SPA shell.

    • React
    • TypeScript
    • Vite
    • Tailwind + Radix
    • Playwright
    • TanStack Query
    Visit site
  • Adaptive HSV Cocolisap Detection running locallyRUNS LOCALLY

    Adaptive HSV Cocolisap Detection

    Undergraduate thesis — computer vision

    My BSCS thesis: detecting coconut scale insect (cocolisap) infestation on coconut leaves from ordinary photographs, without a neural network. The pipeline segments the leaf, builds a MAD-normalised colour-distance map in HSV space, and thresholds it per image rather than against constants tuned once and frozen. Traditional fixed-threshold HSV is implemented alongside it as the baseline, so every claim is a comparison rather than an assertion.

    • Evaluated per pixel on 222 annotated test images against COCO reference masks.
    • Adaptive thresholding lifts recall from 0.15 to 0.64 against the fixed-threshold HSV baseline, and trades precision to get it.
    • Accuracy is not a usable metric on this data: predicting no infestation at all scores 0.8971, because only ~10% of pixels are infested. The work is defended on F1 and IoU instead.
    • Python
    • OpenCV
    • NumPy
    • Tkinter + ttk
    • Matplotlib
    • pandas
    • Pillow
    • threading + queue
    • Roboflow dataset
  • Mercentra Energy Solutions running locallyRUNS LOCALLY

    Mercentra Energy Solutions

    Solar & sustainability — client site

    Marketing and lead-capture site for a solar energy company: solutions catalogue, a solar assessment flow, and financing information.

    • TanStack Start
    • TanStack Router
    • React
    • TypeScript
    • Tailwind + Radix
    • react-hook-form + Zod
    • Framer Motion
    • Cloudflare
  • LIFE OS running locallyRUNS LOCALLY

    LIFE OS

    Personal operating system

    A life-management dashboard that treats goals like a game engine — daily missions, quests, habits and XP feeding character attributes across twelve life domains.

    • React
    • TypeScript
    • Vite
    • Tailwind
    • React Router
    • lucide-react
  • BizPro running locallyRUNS LOCALLY

    BizPro

    Business guidance marketplace

    Server-rendered platform connecting business owners with vetted experts across management, finance, marketing and capital preservation.

    • Node.js
    • Express
    • EJS
    • CORS
    • nodemon
  • BlueTree Domain Selector running locallyNEEDS CONFIG

    BlueTree Domain Selector

    Internal campaign workspace

    Campaign selection tool for vendor inventory: upload an inventory, build campaigns against it, and manage selection config from one workspace.

    Runs locally; the config API needs Supabase credentials.

    • Next.js
    • React
    • TypeScript
    • Supabase
    • xlsx
    • TanStack Query
    • react-hook-form + Zod
    • Recharts
    • Tailwind + Radix
  • AssistED / Jemma AI running locallyNEEDS CONFIG

    AssistED / Jemma AI

    Hackathon — education platform

    Learning platform for Filipino workers with a lesson catalogue, an AI chatbot assistant and speech recognition, backed by a realtime Express and Socket.IO service.

    Client runs; the realtime server needs a MongoDB instance.

    • React
    • Vite
    • Socket.IO
    • react-speech-recognition
    • Express
    • MongoDB + Mongoose
    • bcrypt
    • express-validator
  • No capture
    NEEDS CONFIG

    OFSS

    Online Feedback & Survey System

    Academic capstone: role-based survey platform with session auth, survey building, response collection and an admin reporting area.

    Not captured — needs PHP, Composer and MySQL, none installed on this machine.

    • Laravel 11
    • PHP 8.2
    • Blade
    • MySQL
    • Vite

How I Build

Five stages, every engagement

  1. 01

    Discover

    Understand the business and bottlenecks.

    I map how the work actually flows today — not how the process document says it does. The goal is finding where time leaks and why.

  2. 02

    Architect

    Design the automation and AI system.

    Data model, integration points, failure modes and human checkpoints get decided before a single node is built.

  3. 03

    Build

    Develop workflows, integrations and AI agents.

    Modular workflows with shared error handling, validated AI steps, and version control so a change is reviewable.

  4. 04

    Deploy

    Launch the system into production.

    Staged rollout with monitoring and alerting from day one, plus documentation and a handover the team can actually use.

  5. 05

    Optimize

    Monitor, improve and scale.

    Run data shows what to tighten. Systems get faster, cheaper and more reliable as they earn more responsibility.

Experience

Where the work happens

  1. AI Automation Specialist

    Bare Funeral Group — Australia

    June 2026 — Present

    Automating parts of the finance function, principally the ticket updates that were being maintained by hand across systems.

    • Finance Operations
    • Ticket Automation
    • Systems Integration

  2. AI Automation Specialist

    Design Assists Partners — Australia

    February 2026 — June 2026

    Automated the CRM and the full path from enquiry through to delivery, including the messages relayed out to contractors at each stage.

    • CRM Automation
    • End-to-End Workflow
    • Contractor Comms

  3. AI Automation Specialist

    Roadside Heroes — Guam

    December 2025 — February 2026

    Built a tow request application and a chatbot that handled finance receipts, alongside automating the surrounding business process.

    • Application Build
    • Chatbot
    • Receipt Processing

  4. Founder / Systems Builder

    Repomanta Digital Lab

    2025 — Present

    Building AI automation systems and digital solutions for businesses. Scoping, architecture, delivery and the documentation that keeps a system maintainable after handover.

    • Client Delivery
    • Systems Design
    • Workflow Architecture

Repomanta Lab

What is on the bench right now

The live engineering log — systems mid-build, ideas being tested, and questions I am still chasing down.

  • BUILDING

    Multi-agent operations desk

    A supervisor agent that delegates to specialised workers — triage, research, drafting — with a shared memory of the account it is working on.

    • AI Agents
    • Orchestration
  • BUILDING

    Self-validating workflow templates

    A reusable workflow library that validates itself against real node schemas before deployment, so a broken build never reaches production.

    • n8n
    • Schema Validation
  • EXPERIMENTING

    Document extraction accuracy harness

    A test bench that scores extraction prompts against a labelled document set, so prompt changes are measured rather than guessed at.

    • Evals
    • Document AI
  • EXPERIMENTING

    Voice-to-workflow triggers

    Kicking off structured business workflows from a spoken instruction, with a confirmation step before anything with consequences runs.

    • Speech
    • Automation
  • RESEARCHING

    Cost-aware model routing

    Routing each step to the cheapest model that still passes its accuracy bar, measured per task rather than chosen once for the whole system.

    • LLM Ops
    • Cost
  • RESEARCHING

    Autonomous system observability

    What monitoring should look like when the thing being monitored makes decisions — beyond uptime, into whether the judgement is still good.

    • Observability
    • Reliability

Work With Me

Have a Problem Worth Automating?

I help businesses design AI-powered systems that eliminate repetitive work, connect disconnected tools, and create scalable operations.

Let's Build Something
  • 01AI Automation Systems
  • 02AI Agents
  • 03Workflow Automation
  • 04CRM Automation
  • 05Lead Generation Systems
  • 06API Integrations
  • 07Business Process Automation

Contact

Initialize Collaboration

Describe the process that is costing you time. I will come back with how it would be architected, what it would connect to, and whether automating it is actually worth doing.

repomanta — new-request

Goes straight to my inbox.