Teaching ML engineering
the way engineers learn
Tensorloom was started by practitioners who found that most AI courses teach tools rather than the reasoning behind them. We write syllabi the way we would want to read them.
Our Background
How Tensorloom started
Tensorloom grew out of a study group in Bangsar that met weekly to work through papers on backpropagation and optimisation. The people in that group were already employed as engineers — some in fintech, some in logistics, one in a research unit at a local university. What they shared was a gap between what they could do in code and what they understood about why the methods worked.
After two years of running sessions informally, the founders decided to structure what they had learned into a course format that could reach more engineers across Malaysia. The first cohort ran in early 2023 with nine participants. The assessment rubric from that cohort is still the basis of the current graded notebook format.
The school does not pursue partnerships with employers for referral arrangements, and does not list companies on its website as endorsements. Where employers have hired from past cohorts, that is a matter between the engineer and the employer.
Mission
What we are trying to do
The school's aim is to give engineers working in Malaysia a course option where the syllabus is legible, the assessment is on work product rather than multiple choice, and the workload estimate is honest before you sign up.
We publish readiness criteria and run a quiz before enrolment in the entry course. If the quiz result suggests the course would be too slow or too fast for where you are now, we say so directly. The school's reputation depends on participants finishing with a clearer understanding of what they built, not just a record of attendance.
The employer cohort programme extends the same principle to teams: the problem the team works on is real, the code review is against the employer's actual repository, and the school takes no responsibility for production systems it has not supervised.
Instructors
The people who teach
All instructors have worked in industry as engineers or researchers. Teaching is a primary role, not a side arrangement.
Amirul Rashid
Lead Instructor — ML Engineering
Previously a data engineer at a logistics platform in KL. Teaches the Python for ML Practice course and co-developed the graded notebook assessment format.
Siti Nabilah
Instructor — Deep Learning Systems
Research background in convolutional architectures and distributed training. Leads the fourteen-week Deep Learning Systems course and designs the lab assessments.
Karthik Venkataraman
Cohort Supervisor — Employer Programmes
Leads scoping workshops and fortnightly code reviews for the employer cohort programme. Previous experience managing ML engineering teams at two Kuala Lumpur-based startups.
Standards
How we run our courses
Published syllabus before enrolment
Every course publishes its full topic list, weekly schedule, and assessment rubric before any payment is requested. No content is revealed only after sign-up.
Data protection under PDPA 2010
Personal data collected during enrolment and assessment is processed under Malaysia's Personal Data Protection Act 2010. Retention periods and access rights are documented in the Privacy Policy.
Assessment on work product
Grading is based on submitted notebooks, lab reports, and documented reproduction results — not timed multiple-choice tests. Rubrics are shared at the start of each module.
Sources cited by name
Where course content references published research, the paper is identified by author and title. Participants can verify claims independently. Nothing is presented as proprietary theory.
Compute sandbox for lab work
The Deep Learning Systems course provides a compute environment for the duration. Access is scoped to course materials. Participants are not required to provision their own cloud resources.
Honest fit assessment
The entry course uses a published readiness quiz. If the result indicates a mismatch — too advanced or too early — the school communicates this plainly rather than taking the enrolment regardless.
About ML Education in Malaysia
Engineering practice, not credential collection
Machine learning has moved from academic specialisation to engineering practice in a relatively short period. Most engineers working with ML today learned on the job, through papers, or through short courses that cover API usage rather than what the methods are doing. The gap between using a library and understanding its behaviour shows up when something fails in ways the library's documentation does not explain.
Tensorloom's courses are designed for people at that gap. The Python for Machine Learning Practice course covers the numerical foundations that make later work in deep learning legible — array semantics, broadcasting, why certain data shapes matter, how to structure a notebook so a collaborator can follow it. The Deep Learning Systems course goes further into the decisions that sit behind published training results: what initialisation does, why normalisation helps in some architectures and not others, how to profile training and find where time is actually going.
The employer cohort programme treats this differently again. A team working on a real internal problem does not need a course that simulates work — they need structured engineering review and a method for documenting what they built. That is what the cohort supervision provides: regular code review, a written architecture assessment at the midpoint, and documentation that the engineering team can use after the engagement closes.
All three programmes are built and run from Bangsar, Kuala Lumpur. Participants across Malaysia attend online. The school does not operate satellite locations.
Read the syllabus. Then decide.
Course details, assessment methods, and weekly workload estimates are published before any commitment. Send an enquiry if you have questions not covered there.
Contact the School