Isabella AI Backend Services
Relevant source files
The following files were used as context for generating this wiki page:
Isabella AI (specifically Isabella Villaseñor IA™) represents the governance and ethical core of the RDM Digital / TAMV Online platform. It is defined as a "conscious institutional entity" rather than a standard probabilistic engine, governed by an "Estatuto de Dignidad" (Dignity Statute). Within the project's architecture, Isabella acts as a sentinel and gatekeeper for critical operations, ensuring transparency, security, and human-in-the-loop (HITL) supervision.
Sources: src/components/rdm/isabella.tsx:7-12, README.md:10-15
Architectural Overview
The Isabella AI backend is implemented primarily through Supabase Edge Functions (Deno) that handle evaluation, chat, and protocol enforcement. It operates at the "Federación Tecnológica" level, providing the cognitive infrastructure for the heptafederated kernel.
Evaluation Logic and Decision Making
The isabella-evaluate and kernel-isabella-evaluate services process incoming requests to determine if they comply with platform security and ethical standards. Decisions are categorized as accept, alert, or reject based on content analysis and action types.
The diagram shows the logical flow of the evaluation services used to monitor and log system activity.
Sources: supabase/functions/isabella-evaluate/index.ts:18-35, supabase/functions/kernel-isabella-evaluate/index.ts:18-35
Key Components and Features
Evaluation Metrics
Isabella prioritizes "care efficiency" over traditional metrics like screen time. This involves measuring how well users resolve issues or understand the system after an interaction.
Sources: src/components/rdm/isabella.tsx:11, src/components/rdm/isabella.tsx:95-100
Governance Standards
| Feature | Description | File Reference |
|---|---|---|
| Mandatory XAI | Every decision generates a traceable reasoning graph with Shapley values. | isabella.tsx |
| HITL Threshold | Automatic output blocks if certainty falls below 95%, triggering human review. | isabella.tsx |
| Anti-engagement | Explicit prohibition of dark patterns or addictive reinforcements. | isabella.tsx |
| PQC Signatures | Evaluation results are signed using a unique hash and an isabella: prefix signature. | isabella-evaluate/index.ts |
Sources: src/components/rdm/isabella.tsx:85-115, supabase/functions/isabella-evaluate/index.ts:34-36
Backend Implementation
Edge Function Logic
The evaluation services strictly control which actions are permitted within the ecosystem. The following actions are explicitly defined in the allowed set:
sync_repospublish_businessexecute_protocolingest_telemetryregister_placefederation_statedashboard_summary
Sources: supabase/functions/isabella-evaluate/index.ts:16
Security and Threat Detection
Isabella performs regex-based pattern matching on request bodies to detect malicious intent. Blocked patterns include:
rat,malware,scrap(e|ing)telegram userbot,reverse shell,vigilancia masiva
If these patterns are detected, the system immediately returns a reject decision.
Sources: supabase/functions/isabella-evaluate/index.ts:24-25
Data Persistence and Ledger Integration
Every evaluation is recorded in two primary tables to ensure immutability and auditability:
protocol_executions: Stores the specific protocol (e.g., EOCT-MSR), the action, the decision, and the payload hash.bookpi_entries: Creates a narrative entry of the decision, including a cryptographic signature.
Technical Summary
Isabella AI Backend Services provide the ethical and security layer for the RDM Digital project. By combining automated threat detection with mandatory human-in-the-loop protocols for high-uncertainty decisions, the system maintains "Sovereign Digital Governance." All operations are cryptographically hashed and logged to the bookpi ledger, ensuring that technical execution never violates the underlying constitutional framework of the project.
Sources: src/components/rdm/isabella.tsx:75-80, README.md:90-100