↳ Kuala Lumpur · Est. 2021
We build courses the way we wished we had been taught.
Plain engineering, stated prerequisites, and no claims about what completing a course will do for your salary.
← Back to HomeHow Tensor Kopi started
Tensor Kopi began at a kopitiam table in Bangsar in 2021. Two engineers — one working in distributed systems, one in computer vision — were comparing notes on how they had actually learned what they knew. Neither had done it through a polished online platform. Both had done it by reading papers, breaking things, getting code reviewed, and asking questions of people who had already been through it.
The existing options for engineers in Malaysia who wanted to get into machine learning were, at the time, either very expensive and heavily marketed, or free and completely unstructured. What was missing was something in between: a course with graded assignments, real feedback, stated prerequisites, and a working engineer on the other end of a question.
The first cohort of the foundations course ran in early 2022 with nine people. Every one of them was a working developer. The feedback sharpened what we kept, what we cut, and what we were not being honest enough about in the effort estimates.
Since then we have run multiple cohorts of all three courses. The structure has changed. The content has been revised several times. The approach has not: teach engineering, review code, state prerequisites, and do not make claims we cannot support.
What we are
- A small school, not a platform
- Run by engineers who still write code
- Based in Kuala Lumpur
- Part-time courses for working developers
- Code review on graded assignments
What we are not
- A certification body
- A recruitment agency
- A bootcamp promising outcomes
- A self-paced video library
The people who run the courses
Three engineers, all of whom still work on ML systems in some form. Bios are short because the work speaks louder.
Azrul Wafi
ML Engineering Lead
Spent eight years in production ML at logistics and fintech companies. Runs the foundations course and reviews code on all deep learning submissions.
Siti Rahmah
Deep Learning Systems
Computer vision researcher turned engineer. Leads the deep learning course and the distributed training modules. Has run GPU clusters at two Malaysian companies.
Kaveh Tan
Residency Mentor
Systems engineer with a background in serving infrastructure and model monitoring. Leads the applied residency and the model risk sessions.
How we run courses
The standards we hold ourselves to, written plainly.
Code review on every submission
A senior engineer reads and comments on your code. Not an automated linter. Not a rubric. A person who has written similar code.
Stated prerequisites before enrolment
Every course lists what it assumes in the first line of every module. A self-check quiz is available before you pay so you can judge for yourself whether you are ready.
Honest effort ranges
We publish ranges, not single numbers. "8 to 10 hours" is based on what students report across cohorts. We do not publish the lower bound as the headline figure.
Data privacy and student records
Student submission data is not used for model training or passed to third parties. Assessment records are kept securely and are available to the student on request.
Reproducibility requirement
The deep learning course requires all results to be reproducible. This means seeded runs, logged hyperparameters, and a checkpoint that a reviewer can actually load.
Written feedback with every grade
Every graded assignment returns written feedback alongside the grade. We do not deliver grades alone. The feedback is specific to your submission, not a template.
Engineering education, described plainly
Tensor Kopi operates on the principle that the most useful thing you can do for someone learning to build AI systems is to be direct about what is hard, what takes time, and what this kind of work actually involves day-to-day. The copy on this site is written by engineers, not by a marketing function. Where something is difficult, we say so. Where an outcome is uncertain, we do not dress it up.
Machine learning engineering in Malaysia in 2025 sits at an interesting point. Local datasets exist and are underused. There is strong engineering talent in the country that has not had a structured entry point into ML specifically. The courses at Tensor Kopi were built for that context: engineers who know how to write software but are new to the numerical and systems work that ML demands.
The kopitiam name is not just branding. The physical kopitiam — a coffee shop where people sit for a long time, think out loud, and argue about how things work — is a reasonable model for how we try to run live sessions and office hours. The goal is not to transmit information efficiently. The goal is to help you understand something well enough to debug it at 11pm when the loss curve goes sideways.
We teach engineering. Nothing in any Tensor Kopi course qualifies anyone to practise in a regulated field. If you need that kind of credential, we can point you to the right direction, but we are not the right provider for it.
Read about the courses, then send us a question.
We will tell you which course fits your current level — or be direct if none of them do right now.
Send an Enquiry