OH

AI & Intelligence

Governed AI capabilities for Al Manara Properties LLC: a provider-agnostic gateway, versioned models and prompts, permission-aware retrieval, document intelligence, explainable recommendations, anomaly detection — all advisory by default and fenced by the human approval boundary.

AI-01AI-02AI-03AI-04 AI-05AI-06AI-07AI-08
Provider-agnostic AI GatewayLive · Phase 2

Provider profiles, routing by data class / region / capability / cost / latency / health, fallback order, quotas and data-boundary rules. Timeout, retry, circuit breaker and non-AI fallback are mandatory.

AI-01

Open screen →

Model, Prompt & Tool RegistryLive · Phase 2

Versioned prompts, model policies, tools and schemas with owners, test results, effective dates, pinned provider snapshots and test → staging → production promotion with rollback.

AI-02

Open screen →

Permission-aware RAG & RetrievalLive · Phase 2

Index sources with permission filters, retrieval traces that prove caller-permission enforcement before model context is built, citations with source versions, and a hard exclusion list.

AI-03

Open screen →

Document Classification & ExtractionLive · Phase 2

Processing queue with per-field confidence, source coordinates and validation errors. Low-confidence or high-risk fields hand off to human maker-checker verification; originals stay authoritative.

AI-04

Open screen →

CRM & Property RecommendationsLive · Phase 2

Explainable lead scoring, next actions, property matches and valuation advice with confidence and main factors. Accept / edit / reject with feedback; out-of-tolerance valuations need review.

AI-05

Open screen →

Finance & Ops Anomaly IntelligencePhase 3

Duplicate invoices, unusual commissions, cash/parking exceptions and operational risk patterns — alerts cite contributing records and never reverse, block or accuse without a human decision.

AI-06

Open screen →

Human Approval & Safe Action BoundaryMVP · P0

Allowlisted, least-privilege, server-validated tools (read-only by default). Proposed actions are previewed before confirmation, executed idempotently, and high-risk decisions require human approval.

AI-07

Open screen →

Evaluation, Safety, Cost & ObservabilityLive · Phase 2

Golden datasets, groundedness and extraction accuracy tests, adversarial suites, plus budgets, rate limits and kill switches per environment / tenant / use case, with full run telemetry.

AI-08

Open screen →

Use-case posture (spec §12.2, Table 18) Workflow & rules configurator
Use casePhaseRequired controlProhibited shortcutStatus
Lead extraction, scoring & next actionPhase 2Consent, explanation, quality/bias review and overrideAutomatic exclusion without reviewActive
Listing copy & translationPhase 2Grounded property facts, disclaimer and approvalInvented features or pricesActive
Document OCR / classificationPhase 2Confidence, coordinates and maker-checkerSilent master-data updateActive
Contract QA & RAGPhase 2Governed content, value/source snapshot, citations, clause rules, legal reviewAutonomous clause acceptance or silent source-data overrideActive
Call transcription / coachingPhase 2/3Recording policy, retention, access and factual groundingUnreviewed employment actionPilot
Valuation / recommendationsPhase 3Data quality, comparable evidence and range/confidenceGuarantee or unexplained valuePlanned
AML anomaly / graph intelligencePhase 3Explainable alert, evidence and investigator reviewAutonomous compliance decisionPlanned
Copilot / tool actionsPhase 3Tool allowlist, preview, least privilege and approvalOpen-ended autonomous agencyPlanned

Static render mockup — no live data. Source: ERP Modern Development Specification v2.0 · Covers: AI-01, AI-02, AI-03, AI-04, AI-05, AI-06, AI-07, AI-08 (module hub).