What clients say
about working with us
These are organisations that have engaged Thoughtium for AI and ML work. Their feedback reflects actual project experiences.
Back to Home40+
Completed engagements
4.8
Average client rating
6+
Years ML practice
MY+
Malaysia & SEA region
Feedback from recent engagements
Ahmad Hazwan
Head of Operations, Logistics Firm · KL
"The scope document was something we hadn't seen from other vendors before — a real description of what we'd get, when, and for how much. The supply chain model they built has been running in our planning workflow for two months now without any major issues. The simulation environment was particularly useful for our ops team to understand the trade-offs."
March 2026 · Supply Chain Modelling
Lim Wei Jing
CTO, Marketplace Platform · Selangor
"We came in with a fairly vague sense of what we needed for content moderation. The taxonomy workshop they ran helped us clarify our own policies before any model was built. The final system handles our Bahasa Malaysia and English content reasonably well, and the escalation workflow has made a real difference to our review team's load."
February 2026 · Content Moderation
Priya Nair
Engineering Lead, SaaS Company · KL
"The MLOps engagement moved us from having models that worked on individual laptops to a system where the whole team can run experiments consistently and push to production without it being a manual process. The pipeline took about six weeks to set up. Documentation was solid — we've been able to onboard a new engineer without needing Thoughtium involved."
March 2026 · MLOps Setup
Zulaikha Rashid
Supply Chain Manager, Retail Group · Penang
"What I appreciated was that they were direct about limitations from the start. Our data wasn't in great shape, and they told us that clearly before the project began — and proposed a data preparation phase that added two weeks but made the model actually useful. Other vendors might have just started building and explained the problem later."
January 2026 · Supply Chain Modelling
Tan Kai Sheng
Product Manager, Community Platform · KL
"We had a fairly specific moderation challenge with Mandarin-language content in our forum. The team took time to understand the policy nuances before starting training data work, which I think made a real difference to the final model's usefulness. The confidence thresholds were calibrated well — our human review queue is about a third of what it was before."
February 2026 · Content Moderation
Siti Farhana
Data Team Lead, Manufacturing · Johor
"The MLOps pipeline work was exactly what our team needed. We had a backlog of models that worked fine in notebooks but had never been deployed. Within seven weeks we had a proper CI/CD process, containerised serving, and monitoring set up across our AWS environment. The handover session was detailed — we didn't feel like we were left alone with something we didn't understand."
March 2026 · MLOps Setup
Project stories in more detail
Challenge
A regional logistics operator was managing delivery routes across twelve depot locations manually. Route planning was taking two to three hours per day for the planning team, and inventory placement across depots wasn't aligned with actual demand patterns by location.
What We Built
A constraint-based routing model integrated with a demand forecasting component trained on 18 months of historical order data. The simulation environment allowed the team to test inventory redistribution scenarios without committing to operational changes.
Outcomes
Route planning process reduced from several hours to under 30 minutes daily. The recommended inventory configuration, when piloted across three depots, reduced stockout incidents by roughly 22% over the following six weeks. The simulation environment remains in active use by their planning team.
Challenge
A growing marketplace was reviewing all user-submitted listings manually, which had become unsustainable as volume increased. They needed to automate screening for prohibited listings, misleading descriptions, and counterfeit indicators across listings submitted in Bahasa Malaysia and English.
What We Built
A multilingual text classifier trained on labelled listing samples, with a taxonomy developed collaboratively with their trust team. Confidence thresholds were calibrated so that only borderline cases were routed to human reviewers. The model was deployed as an API integrated with their listing submission pipeline.
Outcomes
Approximately 70% of submitted listings are now screened and actioned without human review. The human review queue for borderline cases is well-defined and manageable. Their trust team now handles edge cases rather than routine screening volume.
Challenge
An analytics company had three ML models that had been developed by individual engineers working in isolation. Models had no versioning, deployments were manual and undocumented, and there was no monitoring. A team member departure had made one model effectively unmaintainable.
What We Built
An MLOps pipeline on their existing AWS environment: MLflow for experiment tracking and model registry, a CI/CD process for model code review and deployment, containerised serving for all three existing models, and a monitoring dashboard with alerting for model performance degradation.
Outcomes
All three models are now deployed through the same reproducible process. The team can run experiments, compare results, and deploy with confidence. A new engineer was onboarded to the ML codebase within two days using the documentation and runbooks included in the handover.
Professional standing
PDPA 2010 Compliant
Malaysia Personal Data Protection Act
MDeC Digital Services Member
Multimedia Development Corporation
PIKOM Technology Affiliate
ICT Industry Association of Malaysia
Ready to discuss your project?
Get in touch and describe what you're working on. We'll be straightforward about what's feasible and what working with us would involve.
Get in Touch+60 3-2539 8174 · [email protected]