Open to freelance projects

I remove the work nobody should do.

Process improvement and automation across reporting, ecommerce operations, and AI agents. I map how the work actually runs, measure where it loses time, then build the n8n and KNIME flows that take the manual step out for good.

Email me See the work

15+ ETL pipelines built

2,500+ minutes of work analysed

4 systems shipped

3 teams across three sectors

Most processes do not fail at the decision. They fail in the handoffs between people who each assume the other one is already doing it.

01 / Selected work

Four systems, built to remove a step

Every project started with the same question: which step is nobody paid to do?

Agentic AI for an agency

Hiject

Agentic AI LLM Automation

Problem

Research and social media planning ran on manual briefing. The same brief structure was rebuilt by hand for every client, and follow ups lived in someone's head.

Built

An agentic AI setup covering three jobs: research assistance, social media flow automation from brief through to plan, and a personal assistant for reminders and small recurring tasks.

Scope

Research agent, brief to plan automation, personal assistant

Ecommerce automation

Omnicom Media Group Indonesia

n8n Chrome extension JavaScript

Problem

The team repeated the same manual steps every day across marketplace and internal tools. The data existed, but it had to be moved from one place to another by hand.

Built

Mapped the process to separate real work from copy and paste, automated the repetitive parts in n8n, and built a Chrome extension so the team could run and read it from the browser they already worked in.

Scope

Process mapping, n8n workflows, browser extension

KNIME ETL for accounting

PT Tunas Ridean Tbk

KNIME ETL Excel

Problem

Accounting reporting was Excel based and manual, and nobody had measured how much of the workload was simply repetitive processing rather than judgement.

Built

Measured over 2,500 minutes of operational activity to isolate the repetitive part, then built and validated 15+ KNIME ETL pipelines over several months of accounting data. Ran trials with the users so the new process actually landed.

Scope

Workload analysis, ETL design, validation, user trials

Livestream reporting

Live & Go · Emtek

Pivot reporting Shopee TikTok

Problem

Weekly livestream performance was rebuilt by hand in pivot tables for Shopee and TikTok, and scheduled duration did not always match what actually aired.

Built

Automated the weekly pivot reporting, reconciled the schedule against total livestream duration so the numbers held up, and analysed monthly datasets to give a clear view of product performance on both platforms.

Scope

Reporting automation, data reconciliation, monthly analysis

02 / Experience

Where this work happened

Ecommerce Specialist Intern, Data Support

Ecommerce Division

Omnicom Media Group Indonesia

Sep 2026 – present

Data support for the ecommerce division: reporting, data preparation, and the process automation that removes the manual steps between marketplace data and the people who need it.

Data Analyst Intern

Marketing Division

Live & Go

Part of Emtek

Mar 2026 – Sep 2026

Managed weekly livestream performance reporting on Shopee and TikTok, cleaned and analysed monthly product datasets, and replaced manual reporting steps with automation.

Business Transformation & Improvement Intern

Transformation Office

PT Tunas Ridean Tbk

Apr 2025 – Mar 2026

Analysed accounting workloads across 2,500+ minutes of operational activity, built 15+ KNIME ETL pipelines over several months of data, and ran the automation trials that moved the team onto them.

03 / Approach

Measure, redesign, then automate

Measure what actually happens

I start with the process as it runs today, not as the written procedure says it runs. Time per step, volume, and where the rework comes from.

Remove and simplify first

Eliminate, simplify, standardise, then automate, in that order. Automating a broken process just makes the mess move faster.

Build it so it stays built

A workflow nobody was trained on gets abandoned. Every build ships with exception handling, a log, and a trial run with the people who will use it.

04 / About

Process work, not slideware

I work in business improvement, which means most of my time sits between the spreadsheet and the people doing the job. My scope runs across process improvement, business process automation, data analysis, and operational excellence.

The tools change with the problem. What does not change is the order: understand the process, prove where it loses time, fix the process, then automate the part that is left.

Focus Process and automation

Daily tools n8n, KNIME, Python, SQL

Time zone UTC+8

Languages English, Indonesian

05 / FAQ

Questions worth asking

What kind of problems do you take on?

Work that is done by hand every week and should not be: reporting that gets rebuilt from the same sources, data that moves between systems by copy and paste, approvals that wait in an inbox, and briefing that starts from a blank page every time.

Which tools do you work in?

n8n for event driven workflows and system integration, KNIME for ETL and data preparation, Python and SQL when the job needs them, and Chrome extensions when the work happens inside a browser that has no API. I pick the tool from the constraint, not from habit.

Do you do the analysis, or only the build?

Both, and they are the same job. The analysis is what tells you which step is worth automating in the first place. Automating without measuring usually means building the wrong thing very efficiently.

Can you take over a workflow someone else built?

Yes, and the first task is always the same: read it, run it, and write down what happens when a step fails. Workflows that nobody documented are usually one silent error away from a wrong report.

How do you make sure the change sticks?

Trial runs with the people who will actually use it, before the old way is switched off. On the accounting pipelines that meant validating output against months of existing results and watching where users hesitated, because that is where the process would have been abandoned.

How do we start?

Send a short description of the process and where it hurts. A paragraph is enough. No deck needed.

Have a process that costs you a day every week?

Tell me what the process does and where it slows down. I will tell you honestly whether it is worth automating.

Email me