IN DEVELOPMENT
Document Intelligence
Extraction, classification, summarization, and structured outputs over authorized documents.
ParsingClassificationSummarizationStructured output
AI at CTG One
What it does for you today
We use artificial intelligence where it genuinely works today.
Today we mainly use AI to help you find information inside our documentation (CTG Knowledge, still in pilot). The rest of our AI capabilities — agents, advanced automation, decision support — are still in development. Further down you can see the technical detail and the real status of each one.
Platform statusIN DEVELOPMENT
There is not yet a general production model, RAG, or agent runtime promoted to LIVE. Architecture and governance are defined; the general capability remains IN DEVELOPMENT.
Evidence before answers
Factual answers should favor verifiable context, traceable sources, and controls over unsupported fluency.
01 · AI Architecture
How intelligence enters the system.
Authorization occurs before the model. Action occurs after validation, policy, and human oversight where required.
01
Authorized data
Sources bounded by identity, role, and business unit.
02
Ingestion
Documents, events, and internal systems pass through validation.
03
Context
Only the information required for the task.
04
Retrieval
Semantic retrieval where the use case requires it.
05
Model
Selected according to policy, task, cost, and evaluation.
06
Agent + tools
Explicit tools, restricted scope, and action limits.
07
Workflow
Deterministic rules connect inference to processes.
08
Human oversight
Required whenever the risk level demands it.
09
Action + audit
Traceable outcome with logs and sufficient evidence.
LIVE
Identity & policy
AuthRBACRLSTool scopes
BETA
Context & retrieval
Curated ingestionpgvectorSemantic retrievalCitations
IN DEVELOPMENT
Models & agents
LLM accessAgentsTool useFallbacks
IN DEVELOPMENT
Evaluation & audit
DatasetsMetricsHuman reviewLogs
02 · Capabilities
Capabilities with explicit maturity.
A diagram, library, or prompt does not make a capability LIVE. Every block must demonstrate implementation, evaluation, and operating evidence.
PARTIAL
Knowledge Systems
Semantic search, RAG, contextual retrieval, and source-grounded answers.
Semantic retrievalRAGCitationsContext boundaries
ROADMAP
AI Agents
Specialized agents with identity, purpose, tools, permissions, and escalation.
Tool scopesPolicyHuman escalationAudit
PARTIAL
AI Automation
Classification, routing, drafting, alerts, and recommendations inside controlled workflows.
RoutingDraftingAlertsRules
ROADMAP
Decision Support
Analysis and recommendations to support decisions without replacing human authority in sensitive processes.
AnalysisPatternsScenariosRecommendations
ROADMAP
AI Observability
Latency, tokens, cost, errors, tool calls, escalation, and quality by use case.
LatencyCostErrorsQuality
03 · Agent Runtime
An agent is a system, not a chatbot.
The target runtime combines model, context, tools, controlled memory, permissions, policy, evaluation, and human escalation.
Model + context
Tools + permissions
Policy + human escalation
04 · CTG Knowledge
The ecosystem’s first verifiable RAG pilot.
CTG Knowledge already has an authenticated pilot with curated ingestion, pgvector, semantic retrieval, server-side model access, and citation metadata. It remains BETA/PARTIAL and will not be promoted to LIVE until reproducible evaluation and sufficient operating evidence exist.
Citation-First AI
Question → retrieval → ranking → context → generation → citations. For internal knowledge, an untraceable answer should be considered inferior to a grounded answer.
Curated ingestionpgvectorSemantic retrievalCitation metadata
05 · Human Authority
AI proposes. Humans authorize.
Autonomy depends on risk. Financial, legal, identity, and other sensitive actions retain explicit human control.
L1
Low risk
Search, classification, summarization, informational support.
May be automated with controls and traceability.
L2
Moderate risk
Drafting, routing, recommendations, prioritization.
Requires contextual review and action boundaries.
L3
High risk
Finance, identity, legal, payments, sensitive changes.
Requires explicit human authorization.
06 · Governance & Security
Control intelligence before scaling it.
Data boundaries, permissions, tool scopes, prompt injection, privacy, versioning, evaluation, fallbacks, cost, and observability are part of the product, not add-ons.
Data boundaries
Tool scopes
Prompt injection
Audit trail
07 · Evaluation
A capability is not ready because it “seems to work”.
Before promoting a system to LIVE, an evaluation dataset, metrics, review process, comparison, and release criteria must exist.
Evaluation dataset
Quality metrics
Human review
Release gate
08 · Applied AI
AI applied to concrete operating problems.
Use cases remain in development or roadmap until sufficient technical evidence exists for each business unit.
Knowledge
Operations
Customer support
Decision support
09 · CTG One OS
Intelligence is built on the shared platform.
CTG One OS provides identity, data, security, transactions, documents, and integrations. CTG One AI adds context, models, agents, and evaluation within authorized boundaries.
CTG KNOWLEDGE · v0.1 PILOT
CTG One’s first functional RAG now has an operating surface.
Authenticated users can query a curated internal corpus while admins control what enters the system. The pilot remains PARTIAL until production migration, corpus, and evaluation are verified.