What People Found Useful
and What They Did Not Expect
Accounts from learners across all three programmes — what they came in with, what shifted, and what they would note for someone considering the same path.
Back to Home180+
Learners across all programmes
4.7
Average satisfaction rating
3+
Years of programme delivery
6
Fine-tuning cohort runs completed
Learner Accounts
Nur Raihan
Product Manager · Shah Alam
I joined L01 after spending about a year working with an engineering team and never quite knowing what to ask. The framing around tokens and context windows was the part that clicked first — it changed how I write briefs for the team significantly. The glossary in Malay was useful for explaining things to colleagues who prefer Bahasa.
L01 — June 2025
Wei Kang
ML Engineer · Petaling Jaya
The L02 capstone was probably the most useful part. I chose a domain task relevant to a project at work, so the whole ten weeks had concrete direction. The GPU credits removed a barrier I was not expecting to solve so quickly. Code review was specific and written — not a quick verbal remark during a session that I would forget by Tuesday.
L02 — May 2025
Priya Pillai
Data Analyst · Kuala Lumpur
I found L03 after finishing a deep learning course and realising my GitHub was difficult to follow even for me. The session on writing up decisions was the one I needed most — the actual technique decisions, not just what the final numbers were. The mock interview debrief was detailed and noted a communication pattern I was not aware of.
L03 — June 2025
Faiz Azman
Operations Lead · Subang Jaya
I was nervous joining L01 without a technical background. The first session handled that well — it was direct about what the programme does not assume. By week three I had a clearer picture of what language models are bad at, which has been more useful than knowing what they are good at. The evening schedule worked around my job without friction.
L01 — May 2025
Lim Xuan Yi
Software Engineer · Cyberjaya
I took L02 having done some fine-tuning before from tutorials, but without a structured view of when it is and is not the right choice. The honest framing on that question was refreshing — the programme covers when adaptation is the wrong tool as a proper topic, not as a footnote. The mentor sessions were individual which made the time useful.
L02 — April 2025
Siti Nabilah
UX Researcher · Bangsar South
L01 was my first structured exposure to how AI systems work. I had been using them as tools without understanding the underlying mechanisms, which made it hard to explain to stakeholders when something went wrong. Week three — on evaluating output quality and identifying failure modes — was the most directly applicable to my day-to-day work.
L01 — July 2025
Three Learner Journeys in Detail
A product team that could not agree on what the AI tool was doing
The Situation
A product manager at a logistics software company enrolled in L01 after several months of frustrating handoffs with engineers. The team had adopted a language model feature, but the PM's mental model of how it worked was different from what the system could actually do. Specifications were consistently off in ways neither side could diagnose.
What the Programme Covered
Week two on tokens and context windows was where the misalignment became clear. The PM realised that assumptions built into specifications — about what the model would "remember" across a conversation — were not possible given how context windows work. Week three on failure modes gave language for describing mismatches to engineers.
What Changed
Specification quality improved measurably. The PM began scoping prompts differently and including explicit constraints about session boundaries. The collaboration did not become frictionless, but the disagreements became more tractable because both sides were using the same vocabulary for the system's actual behaviour.
An engineer who had run training before but not evaluated it properly
The Situation
An ML engineer had completed one adaptation project from a tutorial series before joining L02. The model ran. The training loss went down. But when the model was tested on domain examples, the results were worse than expected and the engineer had no framework for diagnosing why.
What the Programme Covered
The dataset construction sessions revealed that the supervised fine-tuning data had been assembled without considering distribution — the training set was not representative of the actual task distribution. The evaluation module gave concrete methods for assessing performance before and after adaptation, not just tracking training loss.
What Changed
The capstone project involved rebuilding the adaptation pipeline with proper evaluation at each stage. The mentor sessions provided written feedback on the dataset construction decisions. The engineer left with a clearer sense of what proper evaluation looks like and why the original project had underperformed.
A researcher whose projects were technically sound but hard to follow
The Situation
A researcher finishing an ML postgraduate programme had several solid projects but a GitHub that was difficult to navigate. READMEs described what the code did, not why the approach was chosen or what the alternatives were. A senior engineer reviewing the work had commented that it was hard to assess the reasoning behind the decisions.
What the Programme Covered
The documentation sessions focused specifically on writing that explains decisions — the choice of architecture, why a particular dataset was used, what was tried and rejected. The mock interview sessions surfaced a communication pattern: the researcher was over-explaining methods and under-explaining motivations, which made answers longer without being clearer.
What Changed
Three projects were restructured during the programme. The written debrief from the second mock interview noted a significant improvement in how motivations were communicated. The repository documentation shifted from describing results to explaining the reasoning chain that led to them. The researcher noted that the feedback on the communication pattern was the most unexpected and useful thing from the programme.
Standing and Recognition
MDEC Digital Hub Network
Listed in the MDEC Digital Hub network directory for digital skills providers in the Klang Valley.
Registered in Malaysia
Incorporated under the Companies Act 2016 and operating from a registered address in Ara Damansara, Selangor.
4.7 Average Rating
Collected via post-programme survey across all cohorts from 2022 to July 2025. Ratings from individual sessions available on request.
Questions About the Programmes?
Contact Bayang Labs to ask about the next cohort schedule, discuss whether a programme fits your background, or request further information on any of the topics covered.
Ready to Start at Your Layer?
Contact Bayang Labs with your background and the programme you are considering. We will confirm availability and outline what to expect before you commit.
Contact Us