Modelcraft
Workshop setting representing the Modelcraft teaching environment
PLATE 01

Teaching AI development as a trade, not a trend.

Modelcraft exists because most AI courses describe outcomes rather than process. We describe the work.

Back to Home
PLATE 02

How Modelcraft started

Modelcraft was started in Kuala Lumpur in 2022 by a small group of engineers who had spent several years helping software teams adopt ML systems in production. Teaching came out of those conversations rather than from a decision to start a school: the same questions kept coming up, and written-down answers kept landing better than verbal ones.

The school's structure follows directly from that experience. Work orders instead of slides because a syllabus that describes inputs, method, output and hours is a contract the learner can actually evaluate. Written review instead of grades because the feedback you receive on real work is the most transferable thing we can give you.

Modelcraft does not offer regulated qualifications. It does not make claims about employment or earnings. It offers structured, reviewed practice for developers who want to work with AI systems competently — which is a different thing, and an honest one.

The school is based at Damansara Heights, Kuala Lumpur. Live sessions run in Malaysia time (MYT). All sessions are recorded. The teaching staff are engineers who currently work in the field — the content stays close to practice because that is where the teachers spend their time.

PLATE 03

The people who teach and review

AR

Amir Rashid

Lead Instructor — Systems & Retrieval

Has built and run retrieval systems in production since 2017. Teaches the six-month programme and reviews capstone artefacts line by line.

LX

Li Xin

Instructor — Prompt Engineering & Evaluation

Works on model integration at a software firm in KL. Leads the Prompt and Context Engineering course and writes failure taxonomy material from her own incident log.

SP

Siti Putri

Instructor — Research Methods & Reproduction

Spent three years in an applied research role at a regional tech company. Teaches the paper-reading course and designed the reading protocol participants keep after week four.

PLATE 04

How we hold the work to a standard

Written review on every artefact

Feedback is written, specific and tied to the actual code or document submitted. No automated scoring; a person reads it.

Data and privacy handling

Participant work is stored on access-controlled infrastructure. It is not used for any purpose beyond running the course. Personal contact details are held only as long as needed.

Pre-enrolment conversation

For the six-month programme, there is a short technical conversation before enrolment. This protects your time as well as ours — if the fit is wrong, we say so.

Honest scope documentation

Each course page carries a "left out on purpose" section. If a topic is not covered, it is listed. This prevents people from enrolling on a false premise.

Versioned course materials

Course materials are kept in a repository with explicit changelogs. When content changes, participants on that cohort are notified and given the updated version.

One-to-one office hours

Office hours in the six-month programme are individual slots, not open forums. Questions that deserve a considered answer get one, not a quick chat answer from across the room.

PLATE 05

What it means to teach software craft in Kuala Lumpur

Malaysia's software industry has grown quickly enough that many developers found themselves working with AI tooling before the field had developed a body of teaching material worth reading. The courses at Modelcraft were built to fill a specific gap: not introductions for complete beginners, and not academic papers for researchers, but practical instruction for working engineers who need to understand what they are doing well enough to debug it.

The design of each course follows a few plain principles. Prerequisites are honest. Scope is stated before enrolment. The workload is described in hours, not in adjectives. And the review process exists because the most useful feedback on technical work is written, specific and tied to what you actually produced — not to a rubric designed to produce a score.

Modelcraft does not run large open cohorts. The prompt engineering course caps at twelve seats because that is the number where written review stays meaningful. The six-month programme takes a small number of participants each time so that office hours can remain one-to-one. The structure reflects a view about what makes teaching useful rather than a view about what scales.

PLATE 06

An enquiry is a conversation and nothing more.

If you're not sure which course fits, say so in the message. That's a reasonable thing to ask about.

Send an Enquiry