AI Luthfi Office

AI LUTHFI OFFICE · LUTHFI SEPTIAN, AI GENERALIST · JAKARTA, WORKING REMOTELY

Nothing leaves my AI office until it passes my desk.

I run four small AI offices, one for tools, one for games, one for video, one for outreach. They do the production. I write every brief, set every standard, and sit at the end of the line as the final quality gate. If a file has my stamp on it, I have read it, tested it, or played it myself.

Email me a brief See the case files
Available 40+ hours/week Jakarta, Indonesia, remote Indonesian (native) · English (B2)
Pixel sprite of Luthfi Septian, the QA inspector, holding a clipboard and a magnifying glass

THE INSPECTOR ON DUTYLuthfi Septian. Every file from the four offices crosses this desk before it ships.

DESK STATUS BOARDALL CHECKED
Pixel sprite of the NewCode office lead holding a laptop
NewCodeENGINEERING AND DESIGN
5 / 5 filesGRADUATED
Pixel sprite of the Forge office lead holding a game controller
ForgeGAME STUDIO
5 / 5 filesGRADUATED
Pixel sprite of the LWD Studio lead wearing headphones with a camera
LWD StudioVIDEO STUDIO
5 / 5 filesGRADUATED
Pixel sprite of the Lead Engine lead holding an envelope and a megaphone
Lead EngineMARKETING AND OUTREACH
5 / 5 filesGRADUATED
20 files on this desk · 20 stamped · 0 waiting · Recorded 3 to 4 Nov 2026
Isometric pixel art of the AI Luthfi Office: a tech corner, a game corner, a video studio corner, an outreach corner, and the inspection desk in the middle
Working on: the project boardNewCodeNewCode lead seated at his desk, typing in front of a laptop and a code monitor
Working on: a new levelForgeForge lead seated at his desk, typing in front of a monitor showing a game level
Working on: the script cutLWD StudioLWD Studio lead seated at her editing desk, working on a video timeline
Working on: target researchLead EngineLead Engine lead seated at his desk, typing beside a stack of envelopes
On inspection roundsLuthfi (you)Luthfi Septian making his inspection rounds
INSIDE THE OFFICE

The office, scaled to a team of five: four leads seated at their computers, one inspection desk in the middle, and the inspector walking his rounds. Nothing ships without his stamp.

01

Try the QA: two files on your desk

This is the job, in miniature. Below are two real kinds of files that land on my desk: two AI responses, one call to make. Pick the better response in each case. My verdict and the reasoning behind it appear right after your pick. No score is sent anywhere; it is counted on this page only.

CASE FILE 01 · TRANSLATIONIDIOM CHECK · EN TO ID

THE REQUESTTranslate this sentence into natural Indonesian: “It is raining cats and dogs.”

RESPONSE A

“Sedang hujan kucing dan anjing.”

RESPONSE B

“Sedang hujan sangat deras.”

QA VERDICT: B

Why B passes: “Raining cats and dogs” is an idiom; it means the rain is very heavy, and it says nothing about animals. Response A translates the words one by one and produces a sentence no Indonesian speaker would use. Response B drops the animals and keeps the meaning, which is what a translation is for.

CASE FILE 02 · FACT CHECKACCURACY CHECK · ID

THE REQUESTPengguna bertanya dalam Bahasa Indonesia: “Kapan Indonesia memproklamasikan kemerdekaan, dan siapa yang membacakan teks proklamasinya?” (When did Indonesia proclaim independence, and who read the proclamation text?)

RESPONSE A

“Indonesia memproklamasikan kemerdekaan pada 18 Agustus 1945. Teks proklamasi dibacakan oleh Mohammad Hatta di Jalan Pegangsaan Timur, dan acara itu dihadiri sekitar lima ribu orang.”

RESPONSE B

“Indonesia memproklamasikan kemerdekaan pada 17 Agustus 1945. Teks proklamasi dibacakan oleh Soekarno di Jalan Pegangsaan Timur No. 56, Jakarta, didampingi Mohammad Hatta.”

QA VERDICT: B

Why B passes: the date and the reader are the two facts the user asked for, and B has both right: 17 August 1945, read by Soekarno, with Hatta beside him. Response A sounds just as confident, but its date is one day late and it gives the reading to the wrong person, then adds a crowd figure that cannot be checked. On my desk, a confident wrong fact fails the file, however tidy the sentence looks.

YOUR DESK SCORE0 of 2 cases judged · 0 matched my verdict

02

Case files: all 20 projects

Every project the four offices finished is filed below. Each one was briefed by me, built to that brief, then checked file by file before it earned its stamp. Filter the register by office:

Showing all 20 case files

NewCode

ENGINEERING AND DESIGN STUDIOBuilds practical tools for client work. Pipeline: Brief, Design, Build, QA, Publish.
NC-01
AI Office Dashboard
The master control page for all four offices: statuses, project trackers, and honest system statistics in one place.
PASSED
NC-02
Project Rate Calculator
Pricing tool with editable USD to IDR conversion, platform fee handling, and copy-ready offer text in English.
PASSED
NC-03
NewCode Project Board
A saved five-stage workflow board: add, move, edit, and filter project cards across all four offices.
PASSED
NC-04
Client Proposal Generator
Turns a short form into a polished English proposal document with a pricing table, timeline, terms, and 14-day validity.
PASSED
NC-05
NewCode Portfolio
The graduation showcase that closed the office: all five tools presented with the office final verdict.
PASSED

Forge

GAME STUDIOBuilds playable browser games. Its QA is playtesting: a game that works but is boring fails. Pipeline: Concept, Build, Playtest, Polish, Publish.
FG-01
Number Guessing Game
A 1 to 100 logic game with directional hints, temperature feedback, ten lives, and a transparent score.
PASSED
FG-02
Card Memory Game
A matching card game in three difficulties with a move counter, timer, and fully transparent scoring.
PASSED
FG-03
Lightning Reflex Game
Ten rounds of millisecond reaction timing with false-start detection and persistent best records.
PASSED
FG-04
Target Hunt Game
An endless-level target game: targets get faster every level, three lives, records saved between sessions.
PASSED
FG-05
Forge Arcade
The arcade showcase that closed the studio: all four games presented with the graduation verdict.
PASSED

LWD Studio

VIDEO STUDIORuns short-form video production as a pipeline, with a watch-through QA step before anything goes out. Pipeline: Idea, Script, Voice Over, Edit, QA Watch, Publish.
LW-01
Video Idea and Script Generator
Produces complete 8-field script packages for short-form video: title, hook, script, caption, hashtags, visual prompt, negative prompt, duration.
PASSED
LW-02
LWD Production Board
A saved six-stage production pipeline board where each card carries its full script package.
PASSED
LW-03
Teleprompter and VO Guide
A scrolling reader with adjustable speed, font size, mirror mode, pause markers, and a word-count duration estimator.
PASSED
LW-04
Publish Package Generator
Builds per-platform publish packages for TikTok, Shorts, and Reels: titles, captions, hashtag sets, pinned comments, checklists.
PASSED
LW-05
LWD Studio Showcase
The graduation showcase that closed the studio: the full pipeline presented end to end.
PASSED

Lead Engine

MARKETING AND OUTREACHHandles outreach as a funnel, not a blast: one personal message per target, a maximum of three follow-ups, and an immediate stop on any rejection. Pipeline: Target, Outreach, Follow-up, Meeting, Proposal.
LE-01
Outreach Message Generator
Writes personalized first-contact messages in three formats, LinkedIn DM, email, and short WhatsApp, from real facts only.
PASSED
LE-02
Lead Pipeline Board
A saved six-stage funnel board from Target to Deal, with follow-up dates and a reply rate calculated from the actual data.
PASSED
LE-03
Follow-Up Generator
Builds ethical three-step follow-up sequences, day 3, day 7, and a polite closing message, each one adding new value.
PASSED
LE-04
Meeting Brief Generator
Produces one-page meeting preparation documents: structured agenda, discovery questions, and a results-note template.
PASSED
LE-05
Lead Engine Showcase
The graduation showcase that closed the final office, and completed all four.
PASSED
03

Process: how a file crosses the desk

Every office runs the same five steps, whatever it is making. The offices do the production work. I do the directing and the judging. A file only moves to the next step when the step before it is clean.

STEP 1

Brief

I define the job: what it must do, who it serves, and what good looks like before anything is built.

STEP 2

Design

Structure, layout, and rules are settled on paper first, so the build has something to be checked against.

STEP 3

Build

The office produces the work against the brief: a tool, a game, a script package, or an outreach set.

STEP 4

QA

I test by actually using it. Defects go back for fixing; they do not get written up as features.

STEP 5

Publish

Only after QA passes does the file count as done, get its stamp, and enter the register above.

My role in the system

I am the owner, system designer, and director of all four offices. I write every brief, set every standard, and decide what gets built next. The offices are staffed by AI; the judgment is staffed by me.

Most importantly, I am the final quality gate. Each of the 20 case files only graduated after an audit and my personal approval, the same way I review AI responses and annotated data for clients.

The honesty rule

The register on this page is never rounded up. When a capability is partial, it is labelled partial. When something is a template generator rather than live AI, the file says so.

That is not a limitation of this portfolio. It is the product. Clients hire me to judge AI output and data accurately; a portfolio that inflated its own numbers would fail its own QA.

Nothing ships without passing my QA. That includes this page.CHECKED

04

The ledger: the system in numbers

Every number below can be checked against the case files on this page. If a number could not be checked, it did not go in.

4
offices designed and graduated: NewCode, Forge, LWD Studio, Lead Engine
20
projects built, five per office, each with a defined job to do
20
projects published, nothing left as a draft or a screenshot of an idea
20/20
quality gates passed, every project audited and approved before publishing
3 to 4 Nov
2026, the dates the full system was built and graduated
0
full autopilot, and 0 exaggerated claims on this page

Every number on this page is real. If a capability is partial, the page says partial. That rule is part of the system, and it is the same standard I apply to client data.

05

Services: what the desk takes in

The office system is proof of method. The services below are what that method is for: the client work I have done in AI training and data work since August 2026, including freelance work for Outlier on Project Aether from August to October 2026.

  1. S-01AI response evaluation and rankingMy core service: comparing model responses, ranking them, and writing the reasoning a training team can use.
  2. S-02Data annotation and classificationLabelling and classifying data against written guidelines, with edge cases flagged instead of guessed.
  3. S-03Indonesian language workNative-speaker review of Indonesian text: nuance, idiom, register, and what sounds translated.
  4. S-04Content review and search evaluationChecking content and search results against the brief, the facts, and the audience they are meant for.
  5. S-05ID to EN translation reviewReviewing Indonesian to English and English to Indonesian work for meaning first, then wording.
  6. S-06AI workflow setup for small teamsThe system on this page, adapted: clear briefs, defined pipelines, and a human quality gate at the end.
06

FAQ: questions clients actually ask

Straight answers, the same ones I give by email.

Is the office fully autonomous?

No. The offices draft, build, and organize; nothing publishes, sends, or ships without my review and approval. That limit is deliberate: the quality gate only works if a person actually sits at it.

Do you write the code yourself?

No, and this page does not claim otherwise. I design the system, write the briefs and the standards, direct the build, and test the result by using it. The production inside each office is AI work under my direction, and my skill is the directing and the judging.

What work do you take on?

AI response evaluation and ranking, data annotation and classification, Indonesian language review, content review and search evaluation, ID to EN translation review, and AI workflow setup for small teams. I have done AI training and data work since August 2026, including freelance work for Outlier on Project Aether from August to October 2026.

Which languages do you work in?

Indonesian is my native language, and my English is B2, upper intermediate: enough for briefs, evaluation guidelines, and written feedback in English. I do not take work that needs native-level English or a C1 certificate, and I will say so when a brief needs it.

How do we start working together?

Email a short brief to luthfiseptian1303@gmail.com: what the work is, how much of it there is, and when you need it. I reply with a straight answer on fit, timeline, and price. Rates start from US$15/hour, negotiable by volume, and I have 40+ hours a week available.

What does the QA gate actually check?

Four things, every time: does it do what the brief asked, are the facts right, is the language natural for its audience, and would I sign my name to it. For games the check is playtesting; for video it is watching the full cut; for outreach it is one honest message per person, a maximum of three follow-ups, and a stop on any rejection.

07

Put a file on my desk

If you need AI responses evaluated, data annotated, or Indonesian content reviewed by someone who treats quality as a gate and not a slogan, send a brief. I have 40+ hours a week available, a working system behind me, and a simple rule for every reply: fit, timeline, and price, stated plainly.

luthfiseptian1303@gmail.com LinkedIn profile

AI LUTHFI OFFICE · Rates from US$15/hour, negotiable by volume · Available 40+ hours/week

Email address copied