What Engineers Said After Completing the Programmes
Collected from cohort feedback forms. Ratings and quotes are reproduced as submitted, with minor copy-editing for clarity.
Back to Home4.1
Avg. satisfaction (out of 5)
214
Programme completions
68%
RL course completion rate
31
Fellowship projects published
Data from cohorts January 2022 — June 2025. Satisfaction figure is mean of end-of-cohort survey responses (n=187).
From the Cohort Feedback Forms
Ahmad Hafizuddin
ML Engineer · Kuala Lumpur
Papers Course — I had been reading papers by skipping most of the method section and going straight to the results. This course fixed that. The reproduction exercise on a paper without public code was the most useful thing I did this year. My failed attempt taught me more than the five that worked.
June 2025
Nurul Zahirah
Data Scientist · Petaling Jaya
RL Course — I dropped out of two other online RL courses because the explanation of the evaluation problem was always handwaved. Here it was a full two weeks of material. That said, 12 to 15 hours a week was closer to 16 for me — the estimate should probably be wider. The code review on every assignment was worth it.
May 2025
Wong Kai Liang
Research Engineer · Penang
Fellowship — My fellowship project was the most structured piece of research I have done since finishing my master's. The preregistration step felt bureaucratic at first, but it stopped me from changing my evaluation metric halfway through when the results were not what I wanted. External critique was direct and useful.
June 2025
Priya Ramasamy
Software Engineer · Subang Jaya
Papers Course — Practical and well-paced. The peer review section was not something I was expecting and initially I resented the extra time it took. Looking back, reading two other people's notes on the same paper showed me how differently engineers interpret the same method section. That was genuinely useful.
May 2025
Muhammad Faris
PhD Candidate · Universiti Malaya
RL Course — I was already familiar with the theory from my coursework. The value here was the implementation. Writing three algorithms from their original papers and having the code reviewed line by line is not something a typical PhD programme makes time for. I found bugs in my understanding of importance sampling that I had been carrying for two years.
June 2025
Syafiqah Yusof
ML Platform Engineer · Johor Bahru
Papers Course — The material is solid. Three stars off because the schedule was hard to manage with full-time work and two of the weekly sessions ran past their stated end time. I still got what I came for — I can now read a paper appendix without feeling like I am missing something. But the logistics need work.
April 2025
Three Learner Journeys in Detail
From using frameworks to reading the papers behind them
CHALLENGE
A backend engineer moving into ML work found that tutorials explained how to use a library but not why it was designed the way it was. Attempts to read the original papers produced confusion rather than clarity.
APPROACH
Enrolled in the papers course. Worked through the reading methodology, completed four of the six reproduction exercises successfully, and documented two partial failures with detailed notes on where the reproduction diverged from the paper's claims.
OUTCOME
By week six, able to read a paper's method section and produce a working prototype without using reference code. Total time spent: approximately 52 hours over eight weeks, slightly above the median for the cohort.
"The two reproductions that failed were more useful than the four that succeeded. I now know what to look for when I suspect a paper has left something out."
Building a simulation environment for warehouse routing
CHALLENGE
An engineer at a logistics company wanted to evaluate whether RL was a practical approach for routing in a constrained warehouse environment. She had read blog posts about RL but had never implemented a policy gradient method from scratch.
APPROACH
Completed all 15 assignments and received code review on each. For the simulation environment project, built a simplified warehouse routing environment in Python. Evaluation protocol was submitted for peer critique in week 14.
OUTCOME
Completed the programme with a documented evaluation showing that the RL approach outperformed a greedy baseline on three of five test configurations and underperformed on two. The written assessment included both findings.
"I spent sixteen weeks expecting to confirm that RL was the right tool for this problem. The evaluation showed it was not always. That is actually more useful."
Investigating resubmission rates in offline RL benchmarks
CHALLENGE
A postgraduate student in computer science had a specific question about whether offline RL benchmark evaluation was overfitting to published results. He wanted to investigate this empirically but lacked access to compute and to structured research mentorship.
APPROACH
Proposed the question at the fellowship intake. With supervisor support, preregistered the evaluation protocol, ran experiments using the allocated cluster, and submitted the work for external critique from two reviewers in week 22.
OUTCOME
Work released publicly in June 2025. Results were mixed — the main hypothesis was partially supported for two of four benchmark datasets and not supported for the other two. The report documents both directions. External reviewers requested one revision before release.
"Having to preregister my evaluation criteria was the single most important thing the fellowship imposed on me. Without it, I would have quietly shifted the goalposts when the results came in."
Reach Us Directly
Phone
+60 3 9059 6284Address
Level 15, Menara Prestige, Jalan Pinang, 50450 KL
Office Hours
Mon–Fri 09:00–18:00 MYT
Ready to Start Your Own Entry?
Send us a brief note about your background and which programme you are considering.
Open an Enquiry