Helping Development Teams Own Their Model Operations
We are a small team based in Kuala Lumpur. Our work is straightforward: sit with developers, explain the operational side of ML work, and leave them better equipped to maintain it themselves.
Back to HomeHow Numen Works Came Together
Numen Works started in late 2022 from a practical frustration. A small group of engineers had spent years watching ML projects stall not because the models were wrong, but because the operational layer around them was fragile — inconsistent environments, poorly tracked experiments, and pipelines that nobody but the original author understood.
The founders had backgrounds across software engineering, data infrastructure, and developer tooling. They had all, at some point, inherited a model that ran "on the author's machine" and spent weeks reconstructing the context needed to retrain it. That experience shaped the approach at Numen Works: focus on the habits and structures that make model work reproducible, not just on the models themselves.
We set up in Mont Kiara to be accessible to engineering teams across the Klang Valley, while also working remotely with teams elsewhere in Malaysia. The name refers to the idea that knowledge becomes genuinely useful only when it is active — held by people who can apply and adapt it, not stored in a document that nobody reads.
What We Are Here to Do
Our work is developer enablement, not consulting in the traditional sense. We do not hand over a document and leave. We work alongside teams, explain the reasoning behind every choice, and make sure the people doing the work can maintain it after we are gone.
Core Beliefs
- A pipeline the team understands is worth more than a sophisticated one they cannot explain.
- Operational knowledge should belong to the team, not to an outside party.
- The pace of improvement that sticks is the one the team can sustain.
- Tool choices matter less than the discipline of using them consistently.
Who Works at Numen
Wei Chong Lim
Founder & Lead Practitioner
Nine years in software and data engineering. Leads workshop design and the coaching engagements, with a focus on making operational concepts stick for developers who are new to them.
Suraya Ahmad
MLOps Specialist
Background in ML infrastructure and developer tooling. Works primarily on the Pipeline Build Coaching engagements, helping teams shape training-and-deployment pipelines they can own long-term.
Rajan Nair
Advisory & Documentation
Spent six years in technical writing and engineering workflow design. Handles the practices handbooks in the advisory engagements and makes sure written materials are genuinely usable, not just thorough.
Standards We Hold Ourselves To
Documented Reasoning
Every recommendation we make includes written reasoning. Teams receive notes they can refer back to without needing to call us.
Confidentiality
Client code, architecture, and internal documentation is handled as confidential. Mutual NDA clauses are available and standard for advisory engagements.
Scope Honesty
We scope engagements to what the team can realistically absorb and act on. We do not pad hours or recommend changes that would not make a practical difference.
Reproducible Deliverables
Templates, checklists, and reference pipelines are version-controlled and formatted for re-use. Deliverables work in the team's actual environment, not just in a demo context.
Plain Communication
We do not over-explain, but we also do not hide behind jargon. If something is genuinely complex, we say so and break it down — we do not dress it up to sound simpler than it is.
Data Handling Awareness
When engagements involve review of model training data or outputs, we observe appropriate data handling practices in line with PDPA (Personal Data Protection Act 2010, Malaysia).
Developer Enablement in Machine Learning Operations
Numen Works focuses on the part of ML development that often receives the least attention during the early stages of a project: the operational foundation. Model versioning, reproducible training environments, testing strategies for ML pipelines, and structured deployment processes are not advanced topics — they are fundamentals that pay dividends over the lifetime of a model in production.
Engineering teams in Malaysia are increasingly being asked to move model work from research and prototyping into production systems. The gap between a notebook that produces results and a pipeline that the rest of the team can run, debug, and extend is wider than it first appears. That gap is where our work sits.
Our approach is grounded in the understanding that documentation and structure are not overhead — they are the mechanism by which knowledge transfers from one person to a team, and from the present to the future. A well-maintained practices checklist or a clearly annotated pipeline saves disproportionate time when team members change or models need updating.
We work across the range of common MLOps tooling — experiment tracking, data versioning, orchestration, and deployment — without prescribing a particular stack. The right tooling depends on team size, infrastructure, and the operational maturity the team is starting from. Our role is to help teams make that assessment clearly and then build the habits that make whatever tools they choose useful in practice.
Talk to the Team
If you'd like to discuss where your team's operational practices currently stand and what would be most useful to address, we're glad to have that conversation.
Get in Touch