墨舜 AI Lab
Let intelligence move through models, knowledge, data and workflows — and take form in real applications.
The most natural way into the system
Text, voice and vision share one context — interaction is where the system first understands intent.
Answers with provenance, judgement with evidence
Turning documents, policies, project material and business data into capability that can be retrieved, cited and kept current — and that can remember, compare and explain itself.
From strategy to finished work, not from prompt to fragment
The value of generation is not one image or one paragraph, but a reusable, reviewable, repeatable content workflow.
The step that changes system state
Where a model starts calling tools, writing to systems and advancing a process. The highest-value step, and the one that most needs a boundary: every execution must be traceable and stoppable.
Models, knowledge, tools and people in one working order
An AI system that keeps running is tested by permissions, audit, deployment and how people work with it — not by how a demo went.
Every node states its real status. Scan the whole field or move directly into a detail page.
Let voice, vision, context and action happen in one interface.
MoShun AI Lab's interaction entry point — an exploration of how text, voice, vision, context and tool use can share a single space.
Questions and follow-ups share one thread; changing subject does not mean re-explaining the background.
Continuous spoken dialogue, with input intensity reflected in the visual state of the interface.
The camera turns on only when you enable it; frames are never displayed or stored, and it can be switched off at any time.
Switching modality never drops the session, and never reaches outside the configured knowledge scope.
→ Converging on one session and one knowledge scope
Integration mode: text and live voice share one Mo-Voice MoLing session; the camera/vision channel is still not connected.
This is not a feature list. It is how we judge whether an AI application can leave the demo environment — miss any one of these and the application stalls at "looks workable".
Start from the business problem and decide which parts are worth handing to AI at all.
Turn the idea into something operable with the smallest version that actually runs.
Test against real data and a real process rather than a demo script.
Connect to existing systems, permissions and processes, and fix which steps a human must confirm.
Set deployment shape, data boundary and operations to the organisation’s requirements.
Adjust from what real use exposes, not from the result of a single acceptance test.
Improvement returns to exploration: what one round of real use exposes is where the next round starts.
AI Lab moves from a section of the MoShun Technology site to an independent showcase, reorganised around five capability domains.
Before connecting a production model, we built listening, thinking, retrieving and answering into a visible visual language.
Inside an organisation an answer without provenance cannot be trusted. We put citation ahead of raw retrieval quality.
Workflow automation opens to invited organisations, focused on how exception handling and approval steps behave in practice.
