Our Company
Built for
real operations.
Obsidium was founded with a straightforward conviction: AI is most useful when it's integrated thoughtfully into how an organisation actually works, not bolted on as an afterthought.
← Back to HomeOur Story
From machine rooms to production floors.
Obsidium began as a consulting group in Hong Kong's Causeway Bay district, working directly with operations teams who had promising AI models gathering digital dust — not because the models were wrong, but because the deployment context had never been properly addressed.
We formalised our services around three disciplines that repeatedly surfaced as gaps: getting models to hardware, producing annotated data with domain accuracy, and keeping deployed models honest over time. These aren't glamorous problems. They're the ones that determine whether AI investment turns into operational value.
Today, we work with organisations across manufacturing, retail, logistics, and healthcare — each with different constraints, different team structures, and different definitions of success. We take those differences seriously.
Our Mission
Precision over complexity.
Our mission is to make AI genuinely operational for organisations that can't afford to experiment indefinitely. We focus on practical outcomes, honest timelines, and work that holds up after handoff.
Clarity first
We spend time scoping before we start. Ambiguity in the brief creates problems in the field.
Collaborative by design
Your domain expertise shapes our work. We don't arrive with fixed templates.
Measured outcomes
We define what success looks like at the start, so there's no dispute at the end.
The Team
People who've done this before.
Marcus Tang
Director, AI Engineering
Fifteen years deploying machine learning systems in manufacturing and supply chain contexts across Southeast Asia. Built the edge methodology at the core of our deployment practice.
Sophia Lam
Head of Data Operations
Sophia leads annotation quality and process design. Her background spans computational linguistics and large-scale labeling operations for both commercial and research-grade datasets.
Ravi Kumar
Lead, Model Monitoring
Ravi designs monitoring architectures that speak to both engineers and executives. He previously built internal observability tooling for a regional fintech before joining Obsidium.
How We Work
Quality standards across every engagement.
Data Confidentiality
All client data is handled under strict confidentiality agreements. Annotation staff operate on a need-to-know basis. We support NDA arrangements for all project types.
Multi-Pass QA
Every annotation deliverable passes at least two independent review stages before acceptance. Dispute resolution procedures are defined in the project brief.
Infrastructure Integrity
Edge deployments are tested in environments that mirror your production conditions. We don't ship configurations optimised for lab benchmarks.
Monitoring Accountability
Alert thresholds and escalation procedures are agreed before monitoring systems go live. We document what each alert means and what response it warrants.
Clear Scope Documentation
Every project begins with a written scope document. Changes are tracked and agreed in writing. No surprise additions, no scope creep disputes.
Stakeholder Transparency
We provide regular progress updates in language appropriate to the audience — technical detail for engineers, summary views for decision-makers.
Our Approach
AI as infrastructure, not spectacle.
Obsidium serves organisations that have moved past proof-of-concept curiosity and want AI to function reliably as part of their core operations. Our work spans edge hardware configuration, rigorous data annotation workflows, and sustained model performance tracking — each discipline demanding its own methodology and expertise.
We operate from Hong Kong with deep familiarity with the regulatory environment, infrastructure landscape, and operational patterns common across the region. Our team brings experience from manufacturing, retail, logistics, and healthcare contexts, which informs how we approach each new engagement.
The organisations we work with value measured progress over dramatic claims. We share that preference. Our track record is built on engagements that worked — not case studies that obscure the details.
Work with us