João Coelho
PT EN

Open to remote or hybrid roles

João Coelho.

Systems implementation and automation

I spent years on the side that receives a new system in the middle of live operations. Now I work on the other side: I map the process with the team, connect the tools through APIs and stay with it until the system is used. I’m following a learning track for the technical side, starting with Python and SQL, and I apply it to the projects that are live.

Looking for roles in implementation, technical onboarding or automation operations.

Based in
São Paulo · remote or hybrid
Languages
Native Portuguese, advanced English and Spanish, basic German
Citizenship
Brazilian and Spanish (EU)
Education
International Relations, Marketing emphasis, ESPM

Projects

Also built

About

I’ve worked in communications since 2018: content, community design and, for the last five years, film and video. I’m a partner at Mizē, a production studio in São Paulo, where I handle the brands we work with and coordinate delivery.

In 2026 I started building the company’s own systems. I reorganized our Notion, wrote the client-service procedure, connected payments, contracts and WhatsApp through APIs and shipped the projects above. This is the work I want to keep doing, on an implementation team.

  • Communications since 2018
  • 5 years in film and video
  • Client-facing work with brands
  • Studying Python and SQL

Process and clients

  • Process discovery
  • Onboarding
  • Written procedures
  • Notion
  • ClickUp
  • Jira
  • Google Sheets

Integration and automation

  • REST APIs
  • Webhooks
  • n8n
  • Supabase
  • WhatsApp Cloud API
  • AI agents
  • Whisper

Learning

  • Python
  • SQL

Experience

  1. Jun 2025 – present

    10 minto put a quote together, down from about an hour

    Partner · Mizē

    film production company, São Paulo

    • Main contact for brands such as Schutz, Daki, WhatsApp and Volkswagen: briefing, scope, schedule, approval rounds and delivery.
    • Executive production from concept to delivery, including a Daki campaign with about 40 people on set.
    • Mapped how the company really worked (Jun 2026) and rebuilt our Notion in July: CRM, project pipeline, editorial calendar and the client-service procedure our intern then ran on her own.
    • Automated quotes, briefs and crew planning from the brief: about 15 AI agents by area, behind a single approval queue. A quote that took about an hour now takes around 10 minutes, and nothing irreversible goes out without a person’s OK.
    • Mizē Lab integrations: order, payment, contract and a WhatsApp notice to the client with no manual steps, live since November 2026.
    • Client delivery
    • Process
    • Integrations
    • Automation
  2. Jun 2024 – Aug 2025

    147edits measured in the capacity study

    Video producer and editor · Insider Store

    apparel-tech DTC brand

    • Helped build the in-house video team and coordinated its day-to-day operations, with 5 to 6 editors including freelancers: weekly planning, rituals and reporting, alongside the team’s technical lead.
    • A 7-week capacity study (86 shoots, 147 edits) that set a sustainable load of up to 10 shoots and 30 edits a week and flagged the editing bottleneck.
    • Set up the team’s cadence: a weekly technical check with a time estimate per video, Monday shoot-and-edit planning and an async report tracking every piece by channel.
    • Owned 3 of the team’s 9 quarterly key results and built the schedule reviewed at the OKR meetings.
    • Found that only 7% of edits reused existing footage (10 of 147) and ran a task force that turned archived material into 9 new briefs.
    • Operations
    • Capacity
    • OKRs
    • Performance
  3. Aug 2022 – Aug 2023

    20+events covered

    Communication strategist and visual producer · Silveiratech

    regeneration startup, Portugal

    • Connected entrepreneurs, volunteers and partners and helped design the startup’s community.
    • Facilitated creativity and innovation workshops in Portugal and Belgium.
    • Helped plan and run the communication strategy and wrote the company’s English-language copy.
    • Institutional videos, coverage of 20+ events and a documentary made with GAFF Visuals.
    • Community
    • Facilitation
    • English
  4. 2018 – 2022

    290+cities whose decrees I tracked during Covid

    Communications, community and government relations

    Arcos Dorados, Radar Governamental, Sete, Mundos Possíveis, Sesh

    • Government relations at Arcos Dorados (McDonald’s) during Covid: tracked decrees in 290+ cities and helped draft the restaurant rules document with the São Paulo state government.
    • Research and reporting at Radar Governamental for clients such as Uber and Itaú. Then content strategy and copywriting at Sete, Mundos Possíveis and Sesh.
    • Community
    • Research
    • Writing

2017 – 2020

Education

International Relations · ESPM

B.A. with an emphasis in Marketing, São Paulo

The cases in detail

Honu

Training, food and habits, linked to the care team

2026Version 138 live · not yet commercial

People who train and diet with professional support live between a workout app, the nutritionist’s PDF and WhatsApp screenshots. The professional walks into the appointment without knowing what happened that week.

  • Today screen: daily target, macros and meals, with sample data
  • Voice assistant with suggested questions
  • How it works

    1. Appone-tap log, meal photo, voice
    2. honu-aireads the photo, transcribes audio, answers
    3. Supabasedata in São Paulo, private by default
    4. Consentthe client chooses what the team sees
    5. Pro dashboardwho went quiet, adherence, draft message
    6. Suggestionplan and message go back to the app
  • Chat with the assistant, answering from what the person already ate that day
  • Weekly habits inside a 13-week season
  • Body measurements logged by tapping the map
  • Recipe with calories and macros, sample data
  • Professional dashboard in demo mode, with fictional clients
  • Record of a fictional client: weight, training, calories and follow-up
drag sideways

App screens made for the App Store with sample data; the professional dashboard is in demo mode, with fictional clients.

What I did

  • A phone-first app built for short moments: one-tap logging, AI-read meal photos, habits in seasons and body measurements logged on a body map.
  • A text or voice assistant that answers from the person’s own logs (“what should I have as a snack?”), with audio transcribed on the server. It doesn’t prescribe diets or medication.
  • A table of 1,956 foods (TACO, IBGE and USDA) with the source of every value. Trace amounts and missing data never become zero.
  • On the other side, the professional area: who needs attention, per-client record and an invite with a consent form. The client chooses what to share.
  • Demo mode with a fictional gym, to present to partners without touching the database.
  • The same codebase packaged for iOS and Android, with the account deletion and privacy the stores require.

Stack

  • Vanilla JavaScript
  • Supabase (São Paulo)
  • RLS
  • Edge Functions
  • Vision AI
  • Whisper
  • Capacitor 8
  • Vercel
  • Playwright (Python)

Private code: I walk through the architecture and tests on a call.

Numbers

  • 146Playwright test scripts
  • 1,956foods, every value sourced
  • v138versions shipped

Decision

A message drafted with AI reaches the client labelled “text suggested by Honu and reviewed by” the professional, because of Brazil’s CFN resolution 856/2026. The AI organizes and suggests; a person signs off.

What broke

The regression suite grew past two hours and some tests got stuck on environment dependencies. It now runs in two halves, the 3D test runs on its own, and known failures are logged with their version.

↑ Back to projects

  • Backend
  • Payments
  • Integrations
See it live lab.mizestudios.com

Mizē Lab

From order to contract, no manual work

2026In production since Nov 11, 2026

Mizē Lab sells fixed-price websites to small businesses. Every sale went through WhatsApp: agree on the plan, send the contract, issue the charge, check the payment, then notify the client. Each step depended on someone on the team.

  • Mizē Lab home page, with open pricing and the studio’s own images
  • How it works

    1. Siteorder, acceptance, Pix or card
    2. criar-pedidoprice from the database, never the browser
    3. Asaascustomer and charge via API
    4. asaas-webhooktoken checked, order paid
    5. Resend + WhatsAppcontract by email and notice
    6. Panelsubscription and resends
  • On a phone: home, plans and the three-step order, with no data entered
  • On a phone: before and after, the five steps and the briefing for new clients
  • Work: the Mizē site and studies for other businesses
  • Desktop checkout, step 1 of 3, no data entered
  • Published contract, with the clauses that limit rights at the top
drag sideways

Phone screens come from the published site, opened at phone size. The order form is shown empty.

What I did

  • Three-step checkout on the site itself: details, contract acceptance, Pix or card in up to 6 installments.
  • A Supabase function that validates the order, prices it from the database table and creates the customer and charge in Asaas. The browser never sends a price.
  • An Asaas webhook that checks its token, marks the order as paid and emails the contract. The accepted text is stored and verified by hash.
  • Client notices through the WhatsApp Cloud API, a monthly subscription created from a team panel, automatic due-date reminders.
  • Database closed to the public and automatic deletion of messages older than 12 months, as the privacy policy states.

Stack

  • Supabase
  • Postgres + RLS
  • Edge Functions (Deno)
  • TypeScript
  • Asaas API
  • Webhooks
  • WhatsApp Cloud API
  • Resend
  • Cloudflare Pages
  • Turnstile
  • pg_cron

Private code: I walk through the architecture and tests on a call.

Numbers

  • 30/30outside access attempts blocked in the database test
  • 4/4leaks caught when I opened a rule on purpose
  • 1st Pixreal payment made, contract sent by webhook, refunded

Decision

Asaas automatic advance payments switched off. Its terms require the service to be delivered, and a cancellation within 7 days would turn into a buy-back with a non-refundable fee. Advances stay manual, after that window.

What broke

On the first real Pix, only a partial refund went through (the fee was already gone) and the webhook never got the event. I marked the order by hand, and the next day partial refunds were handled in code.

↑ Back to projects

Fichário Pokémon

A market-priced storefront and a dashboard that reads interest

2026Live since Nov 2026 · sales paused

Selling a collection one card at a time needs a fair price for each card, front and back photos and a record of what is already on hold. Without that, cards go cheap or the same card gets promised twice.

  • Owner dashboard: heat map of interest per card for the week
  • How it works

    1. Photosfront and back of every card
    2. fichario.pycatalog of 925 cards
    3. Liga Pokémonmarket price for the right version
    4. Marginnegotiation rule, offers above R$ 250
    5. Storefrontgenerated from data, 3D spin
    6. Stockon hold and sold, live
    7. Dashboardfunnel, heat map, sources
  • Owner dashboard: visitors, funnel to list sent and scroll depth
  • Owner dashboard: cards seen, last card before leaving, visitors per day and source
  • Public storefront with a featured card, 3D spin and price
  • Catalog with filters by language, finish and price range
  • Highlights ordered by price; offers accepted above R$ 250
drag sideways

Dashboard with real data from Nov 16–22, 2026: anonymous counts only.

What I did

  • Owner dashboard: a funnel (visited, opened a card, picked, sent the list), a per-card heat map, how far each visitor scrolled, how many cards they saw and the last one before leaving, traffic source, abandoned lists and top searches.
  • Anonymous events by design: no name, phone or IP. Tagged links per community group show where each visit came from.
  • A Python pipeline that matches all 925 cards against Liga Pokémon market prices and applies a negotiation-margin rule.
  • Photo processing (front, back and cut-outs for the 3D spin) and storefront generation from the data.
  • Public storefront with live stock: on hold and sold are set in a panel only the owner can change.

Stack

  • Python
  • Cloudflare Pages
  • Cloudflare Functions
  • Supabase
  • JavaScript

Private code: I walk through the architecture and tests on a call.

Numbers

  • 5WhatsApp orders in the first week (Nov 14–20, 2026)
  • R$ 12,280list requested by one buyer, 33 cards
  • R$ 13,800list requested by another, 6 cards

Decision

Sales paused on purpose: part of the collection goes to PSA grading first, because graded cards are worth more. A business call, not a system failure.

What broke

The automatic match got the card version wrong (normal, holo or reverse). In the Nov 13 review, 61 cards came out below market; I opened 59 one by one, with the last 3 months of sales, and fixed the prices.

↑ Back to projects

let’s talk.

For a role or an implementation project: