What makes the difference in an AI engagement
Not all AI projects deliver what they set out to. The reasons are usually the same: poor scoping, domain mismatch, and handovers nobody can maintain. Here's how we approach things differently.
Back to HomeSix things that shape how we work
Domain-Specific Expertise
We don't apply generic AI templates to domain-specific problems. Every engagement begins with understanding how your organisation labels, structures, and queries information.
Transparent Process
Written scope before work begins. Milestone updates throughout. No surprises at delivery. You know what's being built and why at every stage.
Maintainable Deliverables
Every pipeline, graph schema, and training material is documented so your team can maintain and extend it. We're not interested in creating dependency.
Local Understanding
Operating in Hong Kong means we understand the regulatory context, the bilingual document environment, and the operational norms of the professional services sector here.
Outcome-Focused Metrics
Success is defined before work starts. Precision and recall targets, query performance benchmarks, or participant competency levels — not vague notions of "AI readiness".
Direct Communication
Small team means you speak with the people doing the work. No account management layer, no communication delays, no briefing documents that lose nuance in translation.
Professional expertise that goes beyond the model
The AI model is rarely the limiting factor in a data project. What determines quality is how well the problem has been framed, how carefully the domain vocabulary has been codified, and how thoughtfully the output schema has been designed.
Voxelith's team brings together computational linguistics, information architecture, and domain consulting. That breadth allows us to sit at the intersection of technical possibility and operational reality.
- Ontology design experience across legal, finance, and research domains
- Bilingual (English/Cantonese) document processing capability
- Over four years of HK-based professional AI engagements
What expertise looks like in practice
It means spending the first week of a knowledge graph project understanding your taxonomy before touching any code. It means writing extraction rules that account for the way your contracts actually use language, not how a training dataset assumed they would. It means training materials that reference real tools your team will encounter, not abstract demonstrations.
Tools chosen for durability, not novelty
We select tools based on what your team can maintain after we leave. That sometimes means choosing a slightly less cutting-edge approach over one that requires specialised knowledge to sustain. Every technology choice is documented with the reasoning behind it.
Technology and tooling chosen with longevity in mind
AI tooling moves quickly. A pipeline built around a model or framework that's deprecated in eighteen months creates operational risk. We favour composable, well-documented approaches that remain manageable as the landscape shifts.
- Documented technology choices with alternatives considered
- Modular architectures designed for incremental update
- No proprietary lock-in to Voxelith tooling
A service model designed around clarity, not complexity
Many AI engagements grow opaque over time — deliverables become undefined, milestones slip without explanation, and the final output differs from what was discussed at the start. We address this through explicit agreement and regular communication throughout.
- Written scope with explicit success criteria before start
- Milestone check-ins with written progress notes
- Handover session included in every engagement
What you can expect from every engagement
A kickoff call to align on scope and priorities. Milestone updates in writing, not just verbal summaries. A delivery session where we walk your team through what's been built. And written documentation covering not just how it works, but the decisions made along the way and what to do if something changes.
Fixed project fees, not open-ended billing
All services are priced as fixed project fees. You know the cost upfront, without needing to track hours or manage a retainer. If scope changes materially, we discuss it before proceeding — not after billing.
Straightforward pricing with no variable billing
Open-ended consulting arrangements create incentives that don't always align with client outcomes. We prefer fixed scope and fixed fees — it aligns our interests with yours from the start.
- All fees quoted in HKD, inclusive
- Milestone payment structure available
- No hidden costs for documentation or handover
Results defined by what your team can actually use
A technically sound knowledge graph that nobody queries is not a success. An extraction pipeline with 95% recall that your team doesn't trust is not useful. We design outcomes around operational adoption, not just technical metrics.
- Success criteria agreed before work begins
- User acceptance testing included in delivery
- Post-delivery questions answered at no additional cost for 30 days
How we define success for each service
For knowledge graphs: query performance against test cases defined in scope. For text extraction: precision and recall benchmarks agreed upfront, with acceptance testing on held-out documents. For training: participant confidence assessments before and after each session, plus a structured exercise demonstrating practical application.
How Voxelith differs from typical AI consultancies
| Consideration | Typical Providers | Voxelith |
|---|---|---|
| Scope definition | Broad statements of intent, refined later | Written scope with explicit criteria before work starts |
| Pricing model | Time and materials, variable cost | Fixed project fees, agreed upfront |
| Domain knowledge | Generalist approach, domain noted but not deep | Dedicated discovery phase, domain codified explicitly |
| Handover documentation | Basic code comments, if any | Full documentation including design rationale |
| Post-delivery support | New engagement required | 30-day question support included |
| Communication | Account manager → project team | Direct with the people doing the work |
| Data confidentiality | Standard engagement terms | NDA signed before any data is shared |
Distinctive features of working with Voxelith
Bilingual document capability
We work with documents that mix English and Cantonese — a common reality in Hong Kong business contexts that many AI tools handle poorly. Our extraction pipelines are built to manage code-switching and mixed-language records.
Human review workflow integration
For high-stakes extraction work, we design human review queues into the pipeline from the start — not as an afterthought. This makes AI-assisted review practical and auditable, rather than a black box process.
Visual exploration for knowledge graphs
Every knowledge graph engagement includes a visual exploration interface — so the people who requested the graph can actually use it, not just the engineers who built it.
On-premise deployment options
Where data sensitivity requires it, we can design and deploy pipelines that run entirely within your infrastructure — no data leaves your environment. This is discussed and agreed during scoping, not added as a late-stage consideration.
Milestones and recognition
Organisations served
Years in operation
Client satisfaction rate
HK industry awards
HKICT Awards — Finalist
AI & Data Analytics Category — January 2026
ISO/IEC 27001 Aligned
Information security management practices
Hong Kong AI Society — Member
Contributing member since 2022
See how these advantages apply to your situation
A brief conversation usually makes it clear whether there's a genuine fit. No pressure — just a direct discussion.
Get in Touch