AI Models

​

Overview

• GROQ (Llama): Fast text generation, chat, structured data tasks
• Google Gemini: Multimodal tasks (voice, images, PDFs)

​

Available Models

​

Primary Chat Model

export const AVAILABLE_MODELS = [
  {
    name: 'Llama 3.1 8B',
    value: 'llama-3.1-8b-instant',
  },
] as const;

export const DEFAULT_MODEL = AVAILABLE_MODELS[0].value;

• Model: Llama 3.1 8B Instant
• Provider: GROQ
• Purpose: General conversation, equipment queries, maintenance assistance
• Speed: Optimized for low latency

​

Model Configuration by Task

​

Text Generation Tasks (GROQ)

​

Chat Conversation

CHAT: {
  provider: 'GROQ',
  model: 'llama-3.3-70b-versatile',
  temperature: undefined,
}

• File: app/api/chat/route.ts
• Model: Llama 3.3 70B Versatile
• Use case: Main chatbot responses, equipment queries, maintenance help

​

Checklist Generation

CHECKLIST_GENERATION: {
  provider: 'GROQ',
  model: 'llama-3.3-70b-versatile',
  temperature: 0.3,
  maxTokens: 2000,
}

• Temperature: 0.3 (focused, consistent output)
• Max tokens: 2000
• Use case: Generating maintenance checklists for equipment
• Source: app/constants/ai.ts:58-63

​

Activity Summary

ACTIVITY_SUMMARY: {
  provider: 'GROQ',
  model: 'llama-3.3-70b-versatile',
  temperature: 0.4,
  maxTokens: 500,
}

• Temperature: 0.4 (balanced creativity)
• Max tokens: 500
• Use case: Automatic summaries of maintenance activities
• Source: app/constants/ai.ts:71-76

​

Work Order Closeout

WORK_ORDER_CLOSEOUT: {
  provider: 'GROQ',
  model: 'llama-3.3-70b-versatile',
  temperature: 0.3,
  maxTokens: 800,
}

• Temperature: 0.3 (precise documentation)
• Max tokens: 800
• Use case: Generating closeout notes for completed work orders
• Source: app/constants/ai.ts:97-102

​

Data Transformation

DATA_TRANSFORMATION: {
  provider: 'GROQ',
  model: 'llama-3.3-70b-versatile',
  temperature: 0.1,
  maxTokens: 1000,
}

• Temperature: 0.1 (very deterministic)
• Max tokens: 1000
• Use case: Structured data transformations and parsing
• Source: app/constants/ai.ts:84-89

​

Multimodal Tasks (Gemini)

​

Voice Transcription

VOICE_TRANSCRIPTION: {
  provider: 'GEMINI',
  model: 'gemini-2.5-flash-lite',
  temperature: 0,
  maxTokens: 500,
}

• Model: Gemini 2.5 Flash Lite (lightweight)
• Temperature: 0 (exact transcription)
• File: app/actions/voice.ts:50
• Use case: Converting audio to text
• Source: app/constants/ai.ts:114-119

​

Voice Command Parsing

VOICE_COMMAND_PARSING: {
  provider: 'GEMINI',
  model: 'gemini-2.5-flash-lite',
  temperature: 0,
  maxTokens: 300,
}

• Model: Gemini 2.5 Flash Lite
• Temperature: 0 (deterministic parsing)
• File: app/actions/voice.ts:148
• Use case: Parsing voice commands to structured JSON
• Source: app/constants/ai.ts:127-132

​

Image Analysis

IMAGE_ANALYSIS: {
  provider: 'GEMINI',
  model: 'gemini-2.5-flash',
  temperature: 0.2,
  maxTokens: 1000,
}

• Model: Gemini 2.5 Flash (full version for vision)
• Temperature: 0.2 (precise analysis)
• File: app/actions/vision.ts:50
• Use case: Analyzing photos of equipment and parts
• Source: app/constants/ai.ts:140-145

​

PDF Extraction

PDF_EXTRACTION: {
  provider: 'GEMINI',
  model: 'gemini-2.5-flash',
  temperature: 0.1,
  maxTokens: 2000,
}

• Model: Gemini 2.5 Flash
• Temperature: 0.1 (accurate extraction)
• File: app/actions/files.ts:47
• Use case: Extracting content from technical manuals
• Max file size: 10MB
• Source: app/constants/ai.ts:153-158

​

Temperature Guide

​

Performance & Timeouts

export const AI_TIMEOUTS = {
  QUICK: 10000,    // 10s - Command parsing
  NORMAL: 30000,   // 30s - Checklist generation
  LONG: 60000,     // 60s - Data transformations
};

​

Retry Configuration

export const AI_RETRY_CONFIG = {
  MAX_RETRIES: 3,
  BASE_BACKOFF_MS: 1000,
  MAX_BACKOFF_MS: 30000,
};

• Max retries: 3 attempts
• Backoff strategy: Exponential (1s, 2s, 4s)
• Max backoff: 30 seconds

​

Streaming Configuration

export const STREAM_CONFIG = {
  maxDuration: 30,      // seconds
  sendSources: false,
  sendReasoning: false,
};

• Max duration: 30 seconds
• Sources: Not sent (to reduce token usage)
• Reasoning: Not sent (for faster responses)

​

System Prompts

​

Main Chat Prompt

• University context (UNEG)
• Technical terminology glossary
• Tool usage instructions
• Safety guidelines

​

Voice Transcription Prompt

• Strict literal transcription rules
• Technical acronym recognition
• No filtering or interpretation

​

Inventory Analysis Prompt

• Structured JSON extraction
• Technical specification detection
• Condition assessment

​

Technical Acronyms

const ACRONYMS_GLOSSARY = {
  UNEG: 'Universidad Nacional Experimental de Guayana',
  UMA: 'Unidad Manejadora de Aire',
  BCA: 'Bomba Centrífuga de Agua',
  TAB: 'Tablero de Distribución Eléctrica',
  ST: 'Subestación Transformadora',
  AA: 'Aire Acondicionado (Split/Ventana)',
  GIMA: 'Gestión Integral de Mantenimiento y Activos',
  OT: 'Orden de Trabajo',
  MP: 'Mantenimiento Preventivo',
  MC: 'Mantenimiento Correctivo',
};

​

Model Selection Best Practices

​

Customizing Models

1. Locate the configuration in app/constants/ai.ts
2. Update the model, temperature, or maxTokens
3. Test thoroughly before deploying

​

Next Steps