Engineering Sustainable AI Systems
We partner with organizations across Southeast Asia to build machine learning systems that work reliably in production environments, emphasizing maintainable architectures and knowledge transfer.
Back to HomeBuilt on Production Experience
lumen dats was established in 2019 by a group of machine learning engineers who had spent years implementing AI systems for financial services and technology companies in Bangkok. We noticed a consistent pattern: organizations could build impressive prototypes, but struggled with the transition to production-grade systems that needed to run reliably under real-world conditions.
The founding team had worked together at a regional fintech company where we were responsible for maintaining fraud detection models, recommendation systems, and risk assessment tools that processed millions of transactions daily. We learned through direct experience what it takes to keep ML systems running when stakes are high: proper monitoring, version control, automated testing, clear documentation, and operational procedures that non-specialists can follow.
Our approach differs from traditional consulting in a key way. Rather than delivering recommendations and moving on, we work directly alongside your engineering teams during implementation. This ensures knowledge transfer happens naturally through shared work, and your team builds the capability to maintain and evolve the systems after our engagement concludes. We measure success not just by whether the system works, but by whether your team understands why it works and how to fix it when problems arise.
Today we serve mid-sized companies and enterprise organizations primarily in Thailand, with occasional work across Southeast Asia. Our clients span financial services, healthcare technology, e-commerce, logistics, and manufacturing sectors. Most engagements involve teams that have already proven their ML concepts and need help establishing the infrastructure and processes required for sustainable production operations.
Engineering Specialists
Our team consists of engineers with hands-on production ML experience who understand both the technical and operational aspects of running AI systems at scale.
Siriwat Pattanaset
Previously led ML infrastructure at a regional payments platform. Specializes in MLOps architecture, model serving optimization, and establishing monitoring systems for production deployments.
Nattapong Kittisakul
Eight years of experience building conversational systems for customer service and internal operations. Focuses on dialogue design, intent classification, and building interfaces that communicate naturally and respectfully.
Apinya Chaiyasit
Background in distributed systems and computational optimization. Diagnoses performance bottlenecks in ML pipelines and implements targeted improvements in inference latency, throughput, and resource utilization.
Our Operating Principles
We follow structured engineering practices and industry standards to ensure the systems we help build are maintainable, secure, and production-ready.
Data Protection
All engagements follow strict confidentiality protocols. We sign NDAs before accessing systems, use encryption for data in transit and at rest, maintain access logs, and can work with synthetic data when full production access isn't necessary.
Testing Standards
We implement comprehensive testing at multiple levels including unit tests for data processing functions, integration tests for pipeline components, and validation procedures for model performance in production-like environments before deployment.
Documentation Practice
Every engagement produces detailed technical documentation including architecture diagrams, operational runbooks, troubleshooting guides, and API specifications. Documentation is written for your team to use independently after we conclude our work.
Version Control
All code and model artifacts are managed through version control systems with clear commit histories. We establish branching strategies, code review processes, and deployment procedures that align with your existing development workflows.
Performance Metrics
We establish clear performance baselines at the start of engagements and track improvements through quantitative metrics including latency percentiles, throughput rates, error frequencies, and resource costs. Progress is measured against agreed targets.
Monitoring Setup
Production systems require active monitoring to catch issues early. We implement logging, alerting, and dashboard systems that track model performance, data quality, system health, and business metrics relevant to your specific use case.
Team Collaboration
We integrate with your existing team structure and communication channels. Our engineers participate in your standups, code reviews, and planning sessions. This collaborative approach ensures smooth handoff when the engagement concludes.
Compliance Alignment
For clients in regulated industries, we adapt our practices to align with compliance requirements including PDPA for Thai organizations, GDPR for EU data subjects, and sector-specific regulations for financial services and healthcare.
Technical Excellence Through Practical Application
Production Focus Over Academic Novelty
The newest techniques from research papers often aren't appropriate for production systems. We prioritize proven approaches with understood failure modes over experimental methods, choosing reliability and maintainability instead of technical sophistication. When simpler models work adequately, we recommend them over complex architectures that are harder to debug and operate.
Knowledge Transfer as Core Deliverable
Building a system that works is not enough if your team can't maintain it. Every engagement includes structured knowledge transfer through paired programming, documentation review sessions, and operational handoff procedures. We consider our work successful when your engineers can troubleshoot issues and make improvements independently.
Honest Assessment of Limitations
Machine learning has genuine capabilities but also real limitations. We provide frank assessments when AI isn't the appropriate solution for a problem, when simpler approaches would be more effective, or when the required data infrastructure doesn't exist. Setting realistic expectations prevents wasted effort and disappointment.
Measurable Improvements
Vague promises about transformation don't help anyone. We establish concrete metrics at the start of each engagement and track progress quantitatively. Whether measuring inference latency reduction, cost efficiency gains, or accuracy improvements, we provide clear evidence of value delivered through our work.
Work With Our Team
If you're looking for practical engineering support to move your ML systems into production or improve existing deployments, we'd be glad to discuss how we might assist.