OUR COMPANY
Built around careful
AI practice
Mutiara AI exists to help Malaysian organisations deploy AI systems that are well-understood, properly evaluated, and suited to the actual work at hand.
WHO WE ARE
Our story
Mutiara AI was founded in Petaling Jaya by a small group of practitioners who had spent years building and evaluating machine learning systems in financial services and operations environments. The recurring problem we encountered was not a shortage of AI tools — it was the gap between what organisations thought those tools could do and what the systems actually delivered in production.
We started Mutiara AI to work differently. Each engagement begins with a careful examination of the problem, not a proposal for a platform. We take on projects where the scope is narrow enough to deliver something useful, and we write down what the system will not do alongside what it will.
The name reflects where we are and what we do. A mutiara — a pearl — forms gradually, in layers, without shortcuts. That describes how we prefer to build AI systems: methodically, with attention to the material at hand, and with patience for the process.
Mission
To deliver AI systems to Malaysian organisations that are honest about their limitations, tested before deployment, and designed to support rather than obscure human judgement.
Operating principles
- Scope is agreed in writing before any work begins
- Evaluation results are shared with the client, not summarised
- Error rates and failure modes are disclosed explicitly
- Handover documentation is written for the internal team, not for us
Based in Malaysia
We work from Petaling Jaya, Selangor. Our team understands local document formats, regulatory context, and the operational conditions that shape how AI systems need to behave in Malaysian organisations.
THE PEOPLE
Our team
Ahmad Zulkifli
Founding Engineer
Ahmad leads model design and evaluation across all engagements. He has spent eight years building extraction and classification systems in banking operations environments.
Nurul Rashidah
AI Systems Lead
Nurul designs conversational systems and oversees deployment architecture. Her background is in enterprise software integration, with a focus on handover to internal operations teams.
Kai-Wen Loh
Audit and Governance Analyst
Kai-Wen conducts model audits and prepares remediation reports. He previously worked in risk management at a regional insurance firm, where he reviewed algorithmic underwriting systems.
HOW WE WORK
Our quality standards
Evaluation-first delivery
Every system is tested against a representative sample of real data before it is deployed. Evaluation reports are written, not summarised verbally, and are provided to the client as part of the deliverable.
Written scope agreement
The scope of each engagement is documented before any development work begins. Changes to scope are handled through a formal amendment process, not through informal requests.
Data handling protocols
Client document data is processed under a data handling agreement. We do not retain client data after the engagement closes, and we do not use client data for training purposes without explicit written consent.
Documented handover
Every engagement concludes with a handover package that includes system documentation, evaluation results, and maintenance guidance written for the client's internal team — not for a technical audience at Mutiara AI.
Monitoring guidance
We provide guidance on how to monitor system performance after handover, including what signals indicate the model is drifting, how to capture failure cases, and when a re-evaluation is warranted.
Independence on audits
When conducting a model audit, we do not also offer to build the replacement system. Our audit findings are independent of any commercial interest in the remediation work that follows.
OUR EXPERTISE
AI practice grounded in operational reality
Mutiara AI works in the space between machine learning research and daily organisational operations. Our practitioners have direct experience building systems that process contracts, routing enquiries, classifying claims, and extracting structured data from unstructured document inputs. That experience shapes how we scope work: we know which problems respond well to current techniques and which do not, and we say so clearly at the start of each engagement.
Malaysian organisations face particular conditions that affect how AI systems need to be designed. Bilingual documents, diverse document quality, locally specific regulatory formats, and internal workflows that have developed over years of practice all influence what a system needs to handle. We account for these conditions in our design and evaluation work, rather than applying templates built for different environments.
Our audit practice draws on the same operational knowledge. When we review an existing model, we look at the conditions under which it was built — the data, the evaluation, the deployment environment — and assess whether those conditions still hold. Governance is not a separate concern from technical quality; we treat them as connected.
Ready to discuss your project?
We keep our client list small so we can give each engagement proper attention. If your project looks like a reasonable fit, we will say so — and if it does not, we will explain why.
Get in Touch