Skip to main content

Technology

Engineered for the conditions, not the demo.

Noise, accents, interruption, ambiguity, audit and air-gapped networks are the design brief — not edge cases discovered after launch.

Core technology

Six systems that have to agree, every time

A virtual human is only as credible as its least reliable subsystem.

Conversation Engine

Model-agnostic reasoning with enterprise guardrails.

Routing, grounding and policy sit above the model layer — so you can change provider, or run open weights on your own hardware, without rewriting a conversation.

Model strategy
Provider-agnostic routing
Grounding
Mandatory retrieval citation
Guardrails
Deterministic policy layer
Fallback
Graceful human escalation

Voice Engine

Conversation at the speed of speech.

Streaming recognition, incremental generation and predictive synthesis are pipelined, so the avatar begins responding while the sentence is still forming.

Turn latency
Sub-second target
Interruption
Full barge-in support
Acoustics
Far-field, high-noise tuned
Languages
40+ with dialect models

Digital Human Renderer

Presence that survives close inspection.

Lip-sync, gaze and micro-expression are driven by the same signal stream as the speech, so the face and the voice never disagree.

Lip-sync
Viseme-accurate, phoneme-driven
Expression
Micro-expression blend system
Gaze
Presence-aware attention model
Render
GPU-accelerated, edge-capable

Knowledge & Retrieval

Your knowledge. Your perimeter. Cited every time.

Every factual claim is retrieved from your approved corpus with a traceable source. Nothing is answered from model memory, and the index can run with no external egress.

Architecture
Private RAG
Citation
Source-traceable responses
Freshness
Incremental re-indexing
Isolation
Zero external egress option

Agent Orchestration

Conversations that finish the job.

Role-scoped agents hold explicit permissions over your systems. The conversation books the appointment or files the request, and records exactly what it did.

Agents
Role-scoped with explicit permissions
Execution
Transactional workflow steps
Audit
Full replay of decisions and actions
Integration
REST, GraphQL, webhook, message bus

Security & Compliance

Built for environments that get audited.

Residency, retention, role-based access, encryption and complete interaction logging are properties of the platform, not add-ons.

Deployment
Cloud, private cloud or air-gapped
Residency
Region-pinned data storage
Access
SSO, SAML, OIDC, role-based
Logging
Immutable interaction audit trail

System architecture

Four layers between a question and a completed action

Each layer is independently replaceable — which is how the platform survives a change of model, vendor or regulation.

Presence

What the person sees and hears

  • Photoreal avatar
  • Real-time lip-sync
  • Gaze & expression
  • Neural voice

Understanding

What the system comprehends

  • Speech recognition
  • Intent & context
  • Sentiment signals
  • Language detection

Reasoning

How answers are formed and bounded

  • Private RAG
  • Role-based agents
  • Policy guardrails
  • Escalation logic

Action

What it does in your systems

  • Enterprise connectors
  • Workflow execution
  • Audit trail
  • Analytics

Intellectual property

Protected by design, not by obscurity.

The hard part is not making a virtual human talk. It is keeping presence, speech, retrieval and action in agreement under noise, interruption and audit. That coordination layer is what we protect.

Patent pending

Synchronised presence pipeline

One signal stream drives speech, visemes and expression together, so the face and the voice cannot drift apart under load.

Bounded retrieval reasoning

Factual responses are constrained to a private corpus with mandatory source traceability; out-of-scope requests route deterministically to escalation.

Multimodal escalation signalling

Linguistic, acoustic and interaction signals decide when a conversation should reach a human — before the person has to ask.

Deployment-invariant runtime

Identical behaviour in managed cloud, private cloud or a fully air-gapped facility, so compliance posture never changes the product.

Get started

Ready to put AI in front of people?

Deploy an enterprise AI virtual human where your customers already are.

Book a demo

We reply to every enquiry within two business days