AI Features

AIIA embeds AI throughout the audit lifecycle — from planning assistance to evidence analysis, workpaper drafting, and continuous monitoring. All AI features operate under strict governance with RBAC enforcement, human approval gates, and full audit logging.

Architecture

AI Capabilities at a Glance

Capability	Description	Where Used
AI Companion	Conversational assistant for audit questions and drafting	All modules
Evidence Summarization	Auto-summarize uploaded documents and evidence	Fieldwork
Narrative Drafting	Draft workpaper conclusions and report narratives	Fieldwork, Reporting
Historical Retrieval (RAG)	Search past audits, findings, and knowledge base	All modules
AI Diff Review	Intelligent comparison of document versions	Fieldwork
Data Agent	Natural language queries against audit data	Dashboards
Document Intelligence	Upload and analyze documents with AI extraction	Standalone
Agentic AI	Multi-step autonomous tasks with human approval	Advanced
KRI Wizard	AI-guided Key Risk Indicator creation	Monitoring
Risk Suggestion	AI recommends risks based on industry data	Universe
Scope Generation	AI drafts engagement scoping memos	Engagements
Finding Drafting	AI drafts CCCER components from test results	Findings

Governance Principles

Every AI feature in AIIA follows these non-negotiable principles:

1. Suggestions Only — Never Automatic

AI outputs are always suggestions that require explicit human action to apply. No AI output is written to official records without human confirmation.

2. RBAC-Constrained

The AI respects the same RBAC rules as the user. It can only access data the user is authorized to see. Cross-role and cross-organization data leakage is prevented at the retrieval layer.

3. Fully Audited

Every AI interaction generates an AuditLogEvent with:

• User who requested the AI action
• Model and provider used
• Input prompt (sanitized)
• Output generated
• Whether the suggestion was accepted or rejected
• Confidence score
• Source citations

4. Citations Required

All AI outputs include source citations — links to the specific workpapers, findings, or documents that informed the response.

5. Confidence Scoring

Each AI suggestion includes a confidence score (0–100%) indicating how certain the AI is about its recommendation. Low-confidence suggestions are flagged for extra scrutiny.

Provider Configuration

AIIA supports multiple AI providers with hot-swappable configuration:

Provider	Local	Cloud	Key Required
Ollama	✅	❌	❌
Gemini	❌	✅	✅
OpenAI	❌	✅	✅
Anthropic	❌	✅	✅
DeepSeek	❌	✅	✅
Custom	✅	✅	✅

For on-premise / air-gapped deployments, use Ollama with locally-downloaded models for full AI functionality without internet access.

See Provider Configuration → for setup instructions.

Getting Started

1. AI Companion → — Chat interface and usage guide
2. Evidence Summarization → — Summarize documents
3. Narrative Drafting → — Draft workpapers and reports
4. Historical Retrieval → — Search past audits with RAG
5. Data Agent → — Natural language data queries
6. Document Intelligence → — Upload and analyze
7. Agentic AI → — Multi-step autonomous tasks
8. Governance → — RBAC, audit trails, and IPPF compliance
9. Best Practices → — Get the most from AI