AI Services Documentation

🤖 Overview

The AI Service (src/services/aiService.ts) is the central gateway to Generative AI providers (Google Gemini). It includes privacy-first features like PII masking and RAG (Retrieval Augmented Generation) support.

⚙️ Configuration

• Provider: Google Generative AI (via Vercel AI SDK).
• Environment Variable: GEMINI_API_KEY or GOOGLE_GENERATIVE_AI_API_KEY.
• Default Models:
  • Generation: gemini-2.5-flash (Note: Check for model availability, fallback to 1.5-flash if 2.5 is not released).
  • Embeddings: text-embedding-004.

🛡️ Privacy Layer

Before sending any prompt to the AI Provider, the system sanitizes data:

• Function: Privacy.maskSensitiveData(text)
• Behavior: Replaces sensitive patterns (Emails, Phone Numbers, NIK/ID) with placeholders like [EMAIL].
• Location: src/lib/privacy.ts (called by aiService).

🧠 Features

1. Text Generation with Tools

The service supports Function Calling via Vercel AI SDK tool().

• Available Tools:
  • searchDocuments: Queries internal Knowledge Base (RAG).
  • createInvoice: Generates an invoice based on natural language commands.
• Context Awareness: teamId and userId are injected into tool execution for security.

2. Embeddings (RAG)

• Function: generateEmbedding(text)
• Use Case: Converts text into vector arrays for semantic search.
• Note: Input text is also sanitized/masked before embedding.

🚀 Usage Example

import { AiService } from '@/services/aiService';

// 1. Simple Generation
const response = await AiService.generateText(
  "Create an invoice for Budi for $50 for Web Design", 
  "team_id_123", 
  "user_id_456"
);

// 2. Embeddings
const vector = await AiService.generateEmbedding("Project documentation about auth");

⚠️ Best Practices

• Never bypass masking: Always use AiService methods, do not call google-ai-sdk directly in components to ensure Privacy compliance.
• Error Handling: AI calls can timeout. Use try/catch blocks and implement retries for critical background jobs.