---
title: "AI Profiles, Hyperparameter Tuning & Model Policy Engine"
description: "Documentation for AI Profiles"
---

## Table of Contents
1. [Overview & Architecture](#1-overview--architecture)
2. [Business & Operational Significance](#2-business--operational-significance)
3. [🎯 User Roles & Key Capabilities](#3--user-roles--key-capabilities)
4. [Visual Interface & Layout](#4-visual-interface--layout)
5. [Field Reference & Hyperparameter Specification](#5-field-reference--hyperparameter-specification)
6. [Telecom-Specific Hyperparameter Tuning & Domain Prompts](#6-telecom-specific-hyperparameter-tuning--domain-prompts)
7. [Operational Security Hardening & Prompt Safety](#7-operational-security-hardening--prompt-safety)
8. [Verification & Diagnostics](#8-verification--diagnostics)
9. [Model Context Protocol (MCP) AI Integration](#9-model-context-protocol-mcp-ai-integration)
10. [Glossary](#10-glossary)

---

## 1. Overview & Architecture

In **Ring2All SBC**, the **AI Profiles** module governs the runtime intelligence policies, model selections, and cryptographic parameter bounds for all artificial intelligence workflows. While AI Providers define *where* models reside and how they are authenticated, AI Profiles determine *how* models behave—binding specific foundation models (`gpt-4o`, `claude-3-5-sonnet`, `llama3.1:8b`) to discrete telecom operational tasks, injecting specialized domain system prompts, and tuning execution hyperparameters.

```
      ┌────────────────────────────────────────────────────────────┐
      │               UPSTREAM AI PROVIDERS GATEWAY                │
      │        (OpenAI, Anthropic, ElevenLabs, Local Ollama)       │
      └─────────────────────────────┬──────────────────────────────┘
                                    │
                                    ▼
      ┌────────────────────────────────────────────────────────────┐
      │               RING2ALL SBC AI PROFILES ENGINE              │
      │           (Stored in sbc_admin.ai_profiles)                │
      ├────────────────────────────────────────────────────────────┤
      │ • Model Binding (gpt-4o, claude-3-5-sonnet, llama3.1)      │
      │ • Hyperparameters (Temperature: 0.0–1.0, Max Tokens)       │
      │ • Resilience Policies (Timeout ms, Retry Attempts)         │
      │ • Domain System Prompts (SIP Diagnostics, Threat Scoring)  │
      └─────────────────────────────┬──────────────────────────────┘
                                    │
        ┌───────────────────────────┼───────────────────────────┐
        ▼                           ▼                           ▼
┌──────────────────┐        ┌──────────────────┐        ┌──────────────────┐
│  NOC COPILOT     │        │  PERIMETER GUARD │        │  USER ACCOUNT    │
│  (MCP TOOLS)     │        │  (AI THREATS)    │        │  (DEFAULT CHAT)  │
├──────────────────┤        ├──────────────────┤        ├──────────────────┤
│ Temp: 0.3        │        │ Temp: 0.1        │        │ Temp: 0.7        │
│ Kamailio RPC     │        │ Heuristic IP Ban │        │ Multilingual UI  │
│ Ladder Summaries │        │ Toll Fraud Score │        │ Platform Guide   │
└──────────────────┘        └──────────────────┘        └──────────────────┘
```

Profiles are recorded in `sbc_admin.ai_profiles` and referenced orthogonally by administrative users, Model Context Protocol (MCP) agents, and real-time security monitors.

---

## 2. Business & Operational Significance

* **Deterministic Telecom Analysis**: Operational telecom tasks (e.g., diagnosing a 488 Not Acceptable Here or interpreting SDP codec mismatches) require strict determinism. AI Profiles enforce low temperature settings (0.0–0.2), preventing hallucinations and ensuring repeatable diagnostics.
* **Cost & Token Quota Control**: Enforces hard token limits (`max_tokens`) and request timeouts (`timeout_ms`), preventing runaway token consumption or billing spikes during high-volume telemetry ingestion.
* **Task Specialization via Domain System Prompts**: Profiles inject deep telecommunications domain instructions directly into model contexts, instructing agents on Kamailio architecture, RTPEngine MOS thresholds, and SIP RFC 3261 compliance without requiring user prompting.
* **Granular Profile Assignment**: Individual administrative users or security monitoring daemons can be bound to distinct profiles, providing tailored reasoning capacity matched to operational privilege.

---

## 3. 🎯 User Roles & Key Capabilities

| Role | Administrative Permissions | Operational Responsibilities |
| :--- | :--- | :--- |
| **AI Systems Architect** | Full Read & Write | Creates specialized profiles, calibrates temperature and token thresholds, and authors domain prompts. |
| **Security Operations Lead** | Read & Write | Configures deterministic profiles for AI Perimeter Guard and audits threat analysis prompt guidelines. |
| **NOC Telecom Engineer** | Read & Consume | Utilizes assigned AI Profiles via the NOC Copilot to perform ladder trace diagnostics and dispatcher rebalancing. |
| **AI Hyperparameter Tuner / Automation Copilot** | Programmatic Tuning & Calibration | Audits active prompt templates, inspects model hyperparameter bounds, and adjusts temperature and token limits via MCP. |

---

## 4. Visual Interface & Layout

### AI Profiles Inventory Table
The main repository interface provides a structured listing of all defined AI behavioral profiles:

![AI Profiles List View](/screenshots/sbc/admin/ai-profiles/ai-profiles-list.png)

* **Profile Name & Provider**: Displays the profile label and its linked upstream AI Provider.
* **Model & Status**: Highlights the targeted LLM model identifier and active toggle state.
* **Quick Actions**: Edit profile parameters, test TTS voice synthesis, duplicate profile, or delete.

### AI Profile Configuration Form
The creation and editing view manages model parameters, timeouts, and domain prompts:

![AI Profile Configuration Form](/screenshots/sbc/admin/ai-profiles/ai-profiles-form.png)

---

## 5. Field Reference & Hyperparameter Specification

| Field Name | Type | Recommended Range | Description |
| :--- | :---: | :---: | :--- |
| **Profile Name** | `string` | 3–100 chars | Unique, descriptive name identifying the profile's operational role (e.g., `Perimeter Threat Analyst`). |
| **Provider** | `select` | Active Provider | Upstream AI engine supplying the underlying API execution. |
| **Type** | `select` | `general`, `security`, `voice` | Operational category scoping where this profile can be consumed across the platform. |
| **Model** | `string` | Valid model ID | Exact model string recognized by the provider (e.g., `gpt-4o`, `claude-3-5-sonnet-20241022`). |
| **Temperature** | `numeric` | `0.0` to `1.0` | Controls randomness. Lower values (0.1–0.3) yield focused, analytical answers; higher values yield creative prose. |
| **Max Tokens** | `integer` | `512` to `8192` | Maximum number of tokens the model is permitted to generate in a single response. |
| **Timeout (ms)** | `integer` | `5000` to `60000` | Maximum round-trip latency in milliseconds before the SBC aborts the request. |
| **Retry Attempts** | `integer` | `0` to `3` | Number of automated retry attempts executed upon encountering upstream 5xx errors or rate limits. |
| **System Prompt** | `text` | Telecom domain guidelines | Permanent base instructions framing the role, guidelines, and constraints of the AI model. |
| **Default Chatbot** | `boolean` | `true` / `false` | When enabled, designates this profile as the platform-wide default for the interactive NOC Copilot. |
| **Enabled** | `boolean` | `true` / `false` | Global administrative toggle enabling or disabling profile consumption. |

---

## 6. Telecom-Specific Hyperparameter Tuning & Domain Prompts

| Operational Scenario | Recommended Model | Temperature | Max Tokens | System Prompt Focus |
| :--- | :--- | :---: | :---: | :--- |
| **SIP Trace Diagnostics** | `gpt-4o` or `claude-3-5-sonnet` | `0.2` | `2048` | Focus on RFC 3261 compliance, SIP response codes, SDP codec mismatches, and RTP port routing. |
| **Perimeter Threat Scoring** | `claude-3-5-sonnet` | `0.1` | `1024` | Strict mathematical evaluation of SIP scanning patterns (SIPCLI, friendly-scanner), brute-force REGISTERS, and dynamic IP bans. |
| **NOC Copilot Assistant** | `gpt-4o` | `0.4` | `4096` | Interactive problem solving, multi-step Kamailio binrpc troubleshooting, and documentation retrieval. |
| **Sovereign Local Diagnostics** | `llama3.1:8b` (Ollama) | `0.1` | `2048` | Air-gapped CDR log classification and private telecom network troubleshooting. |

---

## 7. Operational Security Hardening & Prompt Safety

1. **System Prompt Immutability**: Domain system prompts are locked against user-session overrides, ensuring agents cannot be jailbroken into executing unapproved telephony operations.
2. **Tool Whitelisting Correlation**: AI Profiles correlate with MCP Tool Roles, ensuring that an agent operating under a specific profile only has access to relevant tools (e.g., a diagnostic profile cannot execute IP bans).
3. **Deterministic Seed Control**: Where supported by upstream providers (OpenAI `seed`), diagnostic profiles set deterministic seeds to reproduce exact analysis trajectories during incident investigations.

---

## 8. Verification & Diagnostics

To audit configured AI Profiles, verify active bindings, and test model responses via CLI:

```bash
# 1. Query all registered AI profiles and their hyperparameter settings
sudo -u postgres psql -d sbc_admin -c "
SELECT p.profile_name, pr.name AS provider, p.model, p.temperature, p.max_tokens, p.status 
FROM ai_profiles p 
JOIN ai_providers pr ON p.provider_id = pr.id;"

# 2. Verify default chatbot profile assignment
sudo -u postgres psql -d sbc_admin -c "
SELECT profile_name, model, is_default_chatbot 
FROM ai_profiles 
WHERE is_default_chatbot = true;"

# 3. Test prompt evaluation latency via SBC backend logs
journalctl -u softswitch-sbc-api -n 50 --no-pager | grep -E '(ai_profile|llm_execution)'
```

---

## 9. Model Context Protocol (MCP) AI Integration

The **AI Profiles** module exposes dedicated tools in the Ring2All SBC Model Context Protocol (MCP) server, allowing autonomous systems and supervisors to inspect runtime model parameters, verify default copilot assignments, and adjust hyperparameter bounds dynamically.

### 9.1 MCP Tool Summary

| Tool Name | Action | Risk Level | Purpose |
| :--- | :--- | :--- | :--- |
| `list_sbc_ai_profiles` | Read | `read` | List all AI profiles with models, temperatures, max tokens, and active states. |
| `get_sbc_ai_profile` | Read | `read` | Retrieve full profile configuration including system prompt instructions and timeouts. |
| `update_sbc_ai_profile` | Write | `operational` | Update model assignment, temperature, max tokens, timeout, or system prompt guidelines. |

### 9.2 Tool Schemas & Input Parameters

#### Schema: `list_sbc_ai_profiles`
```json
{
  "type": "object",
  "properties": {},
  "additionalProperties": false
}
```

#### Schema: `get_sbc_ai_profile`
```json
{
  "type": "object",
  "properties": {
    "id": {
      "type": "number",
      "description": "Unique integer ID of the AI profile"
    }
  },
  "required": ["id"],
  "additionalProperties": false
}
```

#### Schema: `update_sbc_ai_profile`
```json
{
  "type": "object",
  "properties": {
    "id": {
      "type": "number",
      "description": "Unique integer ID of the AI profile to update"
    },
    "model": {
      "type": "string",
      "description": "Target model identifier (e.g., 'gpt-4o', 'claude-3-5-sonnet')"
    },
    "temperature": {
      "type": "number",
      "description": "Sampling temperature between 0.0 (strictly deterministic) and 1.0"
    },
    "maxTokens": {
      "type": "number",
      "description": "Maximum token output limit"
    },
    "timeoutMs": {
      "type": "number",
      "description": "Request execution timeout in milliseconds"
    },
    "systemPrompt": {
      "type": "string",
      "description": "Domain system prompt instructions"
    },
    "isDefaultChatbot": {
      "type": "boolean",
      "description": "Designate this profile as the platform-wide default copilot"
    },
    "status": {
      "type": "boolean",
      "description": "True to activate profile; false to disable"
    }
  },
  "required": ["id"],
  "additionalProperties": false
}
```

### 9.3 Sample Tool Execution Payloads

#### Example 1: Listing All Configured AI Profiles
**Request Payload:**
```json
{
  "tool": "list_sbc_ai_profiles",
  "parameters": {}
}
```

**Response Payload:**
```json
{
  "success": true,
  "data": {
    "total": 2,
    "profiles": [
      {
        "id": 1,
        "uuid": "3e4f5a6b-7c8d-9e0f-1a2b-3c4d5e6f7a8b",
        "profileName": "NOC Copilot Primary",
        "providerId": 1,
        "providerName": "OpenAI Production",
        "model": "gpt-4o",
        "type": "general",
        "temperature": 0.3,
        "maxTokens": 4096,
        "timeoutMs": 30000,
        "isDefaultChatbot": true,
        "status": true,
        "createdAt": "2026-06-15T11:00:00Z"
      },
      {
        "id": 2,
        "uuid": "4f5a6b7c-8d9e-0f1a-2b3c-4d5e6f7a8b9c",
        "profileName": "Perimeter Threat Analyst",
        "providerId": 2,
        "providerName": "Anthropic Reasoning Engine",
        "model": "claude-3-5-sonnet",
        "type": "security",
        "temperature": 0.1,
        "maxTokens": 2048,
        "timeoutMs": 45000,
        "isDefaultChatbot": false,
        "status": true,
        "createdAt": "2026-07-20T15:00:00Z"
      }
    ]
  }
}
```

#### Example 2: Updating Temperature & Timeout on NOC Copilot Profile
**Request Payload:**
```json
{
  "tool": "update_sbc_ai_profile",
  "parameters": {
    "id": 1,
    "temperature": 0.2,
    "timeoutMs": 25000
  }
}
```

**Response Payload:**
```json
{
  "success": true,
  "data": {
    "message": "AI profile 'NOC Copilot Primary' updated successfully",
    "profile": {
      "id": 1,
      "profileName": "NOC Copilot Primary",
      "model": "gpt-4o",
      "temperature": 0.2,
      "maxTokens": 4096,
      "timeoutMs": 25000,
      "isDefaultChatbot": true,
      "status": true
    }
  }
}
```

### 9.4 Bilingual Natural Language Copilot Prompts

#### English Prompts
* *"List all AI profiles and show which one is currently designated as the default NOC Copilot."*
  → Agent invokes `list_sbc_ai_profiles()`.
* *"Inspect AI profile ID 2 to review its system prompt instructions for threat analysis."*
  → Agent invokes `get_sbc_ai_profile({"id": 2})`.
* *"Tune AI profile ID 1 by setting temperature to 0.2 for more deterministic SIP ladder diagnostics."*
  → Agent invokes `update_sbc_ai_profile({"id": 1, "temperature": 0.2})`.

#### Spanish Prompts (Español)
* *"Lista todos los perfiles de IA y muestra cuál está configurado como el Copilot por defecto."*
  → Agente invoca `list_sbc_ai_profiles()`.
* *"Inspecciona el perfil de IA con ID 2 para revisar sus instrucciones de system prompt para análisis de amenazas."*
  → Agente invoca `get_sbc_ai_profile({"id": 2})`.
* *"Calibra el perfil de IA ID 1 ajustando la temperatura a 0.2 para diagnósticos de trazas SIP más deterministas."*
  → Agente invoca `update_sbc_ai_profile({"id": 1, "temperature": 0.2})`.

### 9.5 Enterprise Security & Execution Safeguards

1. **Strict Operational Clearance**: Updating AI profiles or system prompts requires `operational` clearance under the `ai_profiles` tool category.
2. **Boundary Validation**: Values for `temperature` are validated to remain strictly between `0.0` and `1.0`, and `maxTokens` is capped at platform-safe boundaries (128 to 16384).
3. **Audit History Tracking**: Every prompt modification or hyperparameter update is recorded in `sbc_admin.audit_logs`, enabling rollback if unexpected reasoning behaviors emerge.

---

## 10. Glossary

* **Hyperparameter**: A configuration parameter set before model execution that directly dictates its reasoning behavior, length, and randomness.
* **System Prompt**: Foundational contextual instructions prepended to every conversation that define the model's persona, operational rules, and safety boundaries.
* **Temperature**: A numerical scaling factor (typically 0.0 to 1.0) governing the probability distribution of predicted output tokens.
* **Token**: A basic unit of text (roughly 4 characters or 0.75 words) processed and generated by LLM architectures.
* **Model Context Protocol (MCP)**: An open architectural standard allowing AI copilots to programmatically manage and tune reasoning profiles.