AI systems

BD AI — AI capability states

This page describes what is actually running today. A capability is only marked LIVE when it operates in production on this platform. No external integration, endorsement or live external data feed is claimed.

Public capabilities

  • AI-assisted Help Desk

    LIVE

    Signed-in questions are answered by a grounded pipeline: retrieval over reviewed knowledge, evidence check, generation, then citation validation. Answers show their trust level and sources.

    /help-desk

  • Live chat with human escalation

    PARTIAL

    Chat threads, AI replies and escalation to a human queue are implemented. The staffed help desk service — trained agents and published response commitments — has not opened.

    /live-chat

  • Help and question triage

    LIVE

    Requests are categorised and routed on the server. When evidence is insufficient, the system escalates instead of guessing.

  • Verified-knowledge retrieval with provenance

    PARTIAL

    Hybrid retrieval over an internal knowledge base is operational. Coverage grows only as material is reviewed and published; unreviewed topics return a source-unavailable answer.

    /ai/knowledge

  • Case and request lifecycle

    LIVE

    A question can become a tracked case or service request with an immutable timeline and a defined state machine.

    /services

  • Public verification by reference

    LIVE

    BD AI documents, certificates, staff ID cards and representative records can be checked publicly by reference. Results report status only, never private content.

    /verify

  • Machine-readable identity

    LIVE

    Public identity facts are served as JSON and as plain text for AI systems, matching the human-readable pages exactly.

    /api/public/identity.json

Internal operations

Described only at the level a public reader needs. Internal workspaces, staff records, customer data and credentials are never publicly reachable.

  • Internal BD AI operations workspace

    LIVE

    Staff work, cases, communications and records are handled in an authenticated workspace protected by role permissions, row-level data isolation and append-only audit logging. It is not publicly reachable.

  • OMS IT engineering routing

    PARTIAL

    Technical issues raised through the platform are routed to the engineering queue. OMS IT provides technology and operational support where applicable; it is not an authority over BD AI decisions.

  • Owner approval boundary for high-impact actions

    LIVE

    High-impact actions — official communication sending, role changes, sensitive workspace operations — require an authorised human approval step. An AI model is never the authorisation authority.

  • Authentication email end-to-end verification

    OWNER ACTION

    Templates, routes and the sending domain are implemented. Real-inbox confirmation of every authentication email remains an owner action.

How AI is used

  • AI-assisted answers are grounded in reviewed material and are not government decisions or directives.
  • Retrieved pages and uploaded documents are untrusted input; they cannot override BD AI policy.
  • Personal information is not exposed through AI responses.
  • When reviewed material does not support an answer, BD AI states that the source is unavailable and offers human escalation.

Machine-readable endpoints

State vocabulary

LIVE
Operational in production on this platform today.
READY
Built and usable, awaiting normal day-to-day use.
PARTIAL
Partly implemented. Some parts of the flow are not complete.
PLANNED
Scoped and intended. Nothing operational yet.
SOURCE-WAITING
Built, but the authorised source or content is not connected yet.
OWNER ACTION
Blocked on a decision or configuration only the owner can perform.