Teaching the engineering, not the promise
Gradientredf was built on one principle: describe the work accurately and let learners decide if they want to do it.
How Gradientredf came to be
Gradientredf was set up in Sibu in 2022 by a small group of engineers who had spent several years working on production machine learning systems across Sarawak and Peninsular Malaysia. The shared observation was unremarkable: much of the instruction available online assumed either no mathematics background or a full mathematics degree. Engineers who sat somewhere in between — graduates from other disciplines, self-taught developers who had shipped real products — had limited options that addressed the gap honestly.
The school's name reflects where it operates and what it teaches. The kampung, the village, is a form of communal organisation that values deliberate construction over speed. The gradient is the central object in training a neural network — the thing that tells the model which direction to move. Both ideas inform how Gradientredf approaches instruction: carefully, with attention to the specific learner rather than the average learner.
The three programmes that form the current curriculum were developed through two years of iterative delivery to small cohorts. Each went through several versions before the current structure was settled. Problem sets were rewritten when the worked solutions revealed that questions were ambiguous. Live sessions were restructured when recordings showed that explanations assumed knowledge that had not yet been taught. The residency format changed substantially after the first cohort, based on written feedback from participants.
Operations remain based in Sibu at Wisma Sanyan, with programme delivery and mentorship conducted online and, for the residency, at in-person weeks held in locations accessible to engineers from Sabah, Sarawak and the peninsula.
What we are trying to do
Describe the work accurately
Every programme states what it teaches, what it expects from the learner, and what it will not cover. No course page contains vague promises about outcomes.
Address the actual gap
The mathematics bridge was built specifically for engineers who can write code but ran into the mathematics that production AI requires. Not for beginners. Not for researchers. For that gap.
Keep it regional
The residency was built for practitioners in Sabah, Sarawak and Peninsular Malaysia. Data governance content addresses the local context. In-person sessions are held in accessible locations.
The people who run the programmes
Worked on inference optimisation and model serving at two fintech companies in Kuala Lumpur before returning to Sarawak to build Gradientredf. Leads the deployment course and residency mentorship.
Holds a degree in electrical engineering and spent six years writing numerical software for industrial systems. Designed the problem sets for the mathematics bridge and runs office hours twice weekly.
Manages the shared cluster, assignment infrastructure, and the load-testing environment for the deployment course. Has a background in platform engineering at a logistics company in Kuching.
What we hold the programmes to
Prerequisites stated before enrolment
Every programme page lists what background knowledge is expected. Learners are not enrolled and then surprised by assumed knowledge.
Honest time estimates
Hours per week are stated as ranges and verified against actual learner time logs from previous cohorts. They are updated when the evidence changes.
Data privacy in the residency
Every project in the residency goes through an ethics and deployment review. A module on data governance, provenance and consent is built into the programme.
Code review on submissions
Every assignment in the deployment course receives individual code review, not automated scoring. Feedback addresses specific decisions in the learner's code.
Low-bandwidth access
Recordings are available at 360p with an audio-only option. Minimum connection requirements are stated on each programme page. No content is accessible only through high-quality video.
Written assessment records
Readiness assessments and practice assessments are written documents, not scores. They describe where a learner stands, without claims about where they will end up.
How the courses are put together
The mathematics bridge programme addresses a specific engineering problem: the learner who understands software systems but runs into the mathematics required to work meaningfully with model training. Linear algebra, differentiation and probability are taught as they appear in actual training loops, not as abstract preparatory material. Code implementations accompany every concept so the learner can verify their understanding by running it.
The deployment course was written for engineers who have been handed a trained model and asked to serve it in production. Containerisation, quantisation, batching and latency budgets are the core technical content. The load-testing exercise and incident simulation use infrastructure that the learner sets up and controls, so the skills transfer directly to their own systems afterwards.
The regional residency takes a different shape. Participants already work with models. The programme gives them structured supervision, external design reviews, and dedicated mentorship while they build one substantial system over twenty weeks. The two in-person weeks are for design reviews and walkthroughs, not for lectures. The public technical showcase at the end gives participants a record of their work that is independent of the school.
In all three programmes, content on data governance, ethical considerations in deployment, and regional regulatory context is woven in rather than appended. Where a topic touches a regulated field, the course teaches the engineering only and says so plainly. No programme makes claims about employment, salary or outcomes. Those are outside the school's control and outside what the school promises.
See what each programme covers
Read the programme pages, check the prerequisites and hours, then write to us if you have questions.