Most onboarding platforms fail the same way: they digitize a paper process and call it modern. A PDF becomes a slideshow. A classroom becomes a Zoom call. The content changes format but not nature, and new hires still spend their first week clicking through material that was designed for someone else.
Work Spatial was built to solve a different problem. The brief was to create an onboarding MVP that used spatial computing, Generative AI, WebGL, and Web3 together — not as a showcase of technologies, but as a coherent system for helping employees understand where they work, what they do, and why it matters. This post is a straightforward account of how we built it, what we decided, and what anyone commissioning a similar build should know before they start.
Why This Stack, and Why Now
The combination of these four technologies is not arbitrary. Each one solves a specific problem that the others cannot.
WebGL solves the access problem. Enterprise IT environments are restrictive. Getting an app approved, installed, and maintained across hundreds of workstations is a project in itself. WebGL runs in the browser — no installation, no IT ticket, no device dependency. When the client asked how they could deploy an immersive experience to a distributed workforce, the answer was not a native app. It was a browser-based 3D environment built on WebGL. The same reasoning drove our work on NBK Virtugate, where employees could explore a virtual bank environment without touching their IT setup.
Spatial computing solves the presence problem. There is a measurable difference between reading about a workspace and moving through one. Research shows that learners retain 75% more information in immersive environments than through traditional methods. That gap exists because presence — the subjective sense of actually being somewhere — activates different cognitive processes. Spatial computing gives onboarding physical logic: here is where the team sits, here is the workflow, here is the escalation path visualized as something you can walk through rather than something you read about.
Generative AI solves the personalization problem. A new hire in a technical role and a new hire in a client-facing role should not receive identical onboarding. Generative AI allows the platform to adjust content, pacing, and guidance based on role, prior experience, and early engagement signals. It also allows the system to surface answers to contextual questions without requiring an HR team member on call. Critically, this is not AI replacing human connection — it is AI removing administrative friction so human connection can happen at the right moments.
Web3 solves the ownership problem. Credentials, completion records, and certifications stored on-chain belong to the employee, not just the company's HRIS. For enterprises thinking about portability, verification, and trust across organizational boundaries, this matters. For the MVP, the implementation was intentionally lightweight — we abstracted wallet complexity entirely from the user interface. New hires signed up with email and password. The wallet was created on the backend. This is the right call: research consistently shows that forcing Web3 UX concepts like seed phrases onto mainstream users at the point of signup kills adoption before it starts.
The Architecture Decisions That Actually Mattered
Start with the delivery layer
The first decision was WebGL over native. This is worth stating plainly because it is still not the default instinct for many enterprise buyers who associate 3D experiences with headsets or installed applications. For Work Spatial, the target was a distributed workforce accessing onboarding through corporate devices. WebGL was the only viable path.
We evaluated framework options carefully. The tradeoffs between Three.js, Needle, and server-side rendering approaches like Unreal's pixel streaming are real. Pixel streaming offloads rendering to a server and streams the result, which sounds attractive but introduces network latency that degrades interactive experiences. Three.js gives developers precise control but loads time depends entirely on how well assets are managed. Needle offered smart asset optimization with automatic level-of-detail generation and explicit support for spatial computing on visionOS — a consideration for forward compatibility.
For an MVP, the right call is usually the framework your team knows well enough to move fast without creating technical debt that blocks iteration. We built on foundations we have used across multiple projects, including our WebGL work for NBK, which meant we could move quickly on the core experience while keeping the codebase clean.
Edge matters for presence
One of the less-discussed requirements for spatial computing experiences is latency. For head-tracked environments, response times above 20 milliseconds start to break presence and can cause discomfort. Cloud-only architectures often introduce 50–200 milliseconds of round-trip latency, which is fine for video conferencing but problematic for interactive 3D. Edge computing can reduce this to 5–10 milliseconds by processing data close to the user.
For Work Spatial, we designed the architecture with this constraint in mind. Not every interaction requires edge processing — content delivery, for example, tolerates higher latency. But spatial tracking and interactive AI responses need to feel immediate. Separating these two categories in the architecture prevents a single latency problem from degrading the entire experience.
The AI integration layer
We used Generative AI at two points in the platform: personalized content sequencing and a conversational guide that answered role-specific questions in context.
The content sequencing component is the safer of the two. It adjusts what a new hire sees and when, based on their role profile and completion signals. This is well-understood territory — recommendation logic applied to onboarding content — and it delivers measurable value without introducing the consistency risks that come with generative outputs in compliance-sensitive contexts.
The conversational guide required more careful scoping. A frequent mistake in enterprise AI builds is using large language models for tasks that require determinism. Compliance information, policy specifics, and contractual terms need consistent, accurate answers — not probabilistic generation. We scoped the conversational AI to orientation content: culture, team structure, workflow context, and FAQs. For anything requiring policy-level accuracy, the guide surfaced static verified content rather than generating a response.
This is the architectural principle that enterprise teams need to internalize before they commission a build: map where variation is acceptable versus where it is not, and use generative AI only in the former category.
The Web3 layer at MVP scale
We did not build a full decentralized identity system for the Work Spatial MVP. That would have been the wrong scope. What we built was a lightweight credentials layer: completion records written to chain, accessible and portable, but invisible to the user during normal operation.
The three implementation options for Web3 in enterprise contexts are: build in-house, fully outsource, or integrate a specialized platform. Building in-house carries the highest security risk — cryptographic systems are attack surfaces that require deep expertise to secure properly. Full outsourcing removes that risk but introduces counterparty dependency. For Work Spatial, we integrated a platform layer that handled wallet creation and key management while giving the client full control over the user experience. This is the approach we would recommend for any enterprise client exploring Web3 credentials for the first time.
What We Learned Building It
The MVP scope trap is real
The temptation to build a "small version of the full product" rather than a focused test of a core value proposition cost several weeks on early iterations. When you are combining four technology layers, each one has its own backlog of desirable features. Spatial audio. Avatar customization. Branching narrative. Advanced analytics. These are all worth building eventually. They are not all worth building in the MVP.
We applied a strict prioritization filter: what is the one thing this MVP needs to prove? For Work Spatial, it was that a new hire could complete a meaningful onboarding journey in a spatial environment, with AI personalization, accessible from a browser, and receive a verifiable completion credential. Everything else was deferred. That clarity made the build faster, the feedback cleaner, and the iteration cycle shorter.
The client's response reflected this: "Mohamed and Mazen were very instrumental in helping us develop our MVP. Plethora of knowledge, very professional, work great to keep a project on schedule. We recommend them highly."
Shipping on schedule is not incidental. It is the result of not overbuilding.
Presence requires design, not just technology
The research on learning retention in VR environments contains an important caveat: the retention advantage only appears when users report high presence — the subjective sense of actually being in the environment. Participants with low presence showed no retention benefit over conventional learning, regardless of the technology used. This means that immersive technology does not automatically produce immersive experience. The spatial design, the interaction model, the audio environment, and the pacing of the experience all contribute to whether a user feels present or just feels like they are operating a complicated interface.
In our work across projects — from the Iman VR reconstruction of historical environments to the Empathy Lab VR training scenarios for UK rail staff — this principle holds consistently. Technical fidelity matters less than experiential coherence. A spatially logical, well-paced environment with moderate visual quality will outperform a technically impressive environment with poor interaction design.
The organizational change problem is not optional
The most common failure mode in enterprise AI and XR builds is not technical. It is organizational. Teams closest to the problem need to own the solution, but without governance, parallel efforts create incompatible systems. For Work Spatial, we worked embedded within the client's pipeline from early architecture through launch. This is our standard model, not a special arrangement. The embedded approach means we catch integration problems early, understand the organizational constraints that shape what is actually deployable, and avoid building something that works in isolation but fails in context.
A Pre-Build Checklist for Enterprise Teams
If you are scoping an AI-driven XR onboarding platform, these are the questions to resolve before the first line of code is written:
Access and deployment
- [ ] What devices will employees use, and is native installation feasible at scale?
- [ ] Does your IT environment support WebGL in corporate browsers?
- [ ] What is the maximum acceptable load time on your network?
AI scope
- [ ] Which parts of the onboarding content require consistent, verified answers?
- [ ] Which parts tolerate personalized, generated responses?
- [ ] Who owns content accuracy and updates for AI-surfaced information?
Spatial experience
- [ ] Have you defined what "presence" means for your specific use case?
- [ ] Is the spatial environment organized around actual workflow logic, or just visual novelty?
- [ ] What is the latency tolerance for interactive elements?
Web3 layer
- [ ] Are you implementing Web3 for credentials, for ownership, or for both?
- [ ] Have you abstracted wallet complexity entirely from the user-facing experience?
- [ ] Who manages key custody, and what happens if a provider relationship ends?
MVP scope
- [ ] What is the single hypothesis this build needs to validate?
- [ ] Which features are deferred, explicitly, and documented as post-MVP?
- [ ] What does success look like at 30, 60, and 90 days post-launch?
The technology stack for AI-driven XR onboarding is mature enough to build on. The failure points are almost always in scope definition, organizational ownership, and architectural decisions made too quickly in the early weeks. Get those right, and the build follows.