Content Generator with Gemini AI

A TypeScript application that uses Google's Gemini AI to generate content based on configurable goals and guidelines, with automatic evaluation and retry logic to ensure quality.

Features

• PromptOrchestrator: Converts goals and guidelines into structured LLM prompts
• GeminiService: Handles communication with Google's Gemini AI API with unlimited token generation
• Evaluator: Automatically evaluates generated content against guidelines using AI
• Smart Retry System: Automatically retries generation with enhanced prompts when content doesn't meet quality standards
• Manual Score Calculation: Calculates scores based on percentage of guidelines passed
• TypeScript: Full type safety and modern development experience

Requirements

• Node.js 16+
• TypeScript 5+
• Valid Gemini API key

Setup

1. Install dependencies:

   npm install

2. Get your Gemini API key:

   • Visit Google AI Studio
   • Create a new API key
   • Copy the key

3. Set up environment variables: Create a .env file in the project root:

   GEMINI_API_KEY=your_actual_api_key_here

4. Run the application:

   npm start

Usage

The application will:

1. Generate a structured prompt from your goal and guidelines
2. Send the prompt to Gemini AI for content generation
3. Automatically evaluate the generated content against all guidelines
4. Calculate a score based on the percentage of guidelines that passed
5. If the score meets the threshold (100% by default), display the accepted content
6. If not, retry with enhanced prompts focusing on failed guidelines (up to 3 attempts)
7. Display the best result if all attempts fail to meet the threshold

Configuration

You can modify the following in main.ts:

• Goal: The main objective for content generation (line 8)
• Guidelines: Array of requirements the content must meet (lines 10-18)
• MAX_ATTEMPTS: Number of retry attempts (default: 3)
• SCORE_THRESHOLD: Minimum score percentage to accept content (default: 100%)

API Reference

PromptOrchestrator

• generatePrompt(failedGuidelines?): Creates a formatted LLM prompt, optionally emphasizing previously failed guidelines
• setGoal(goal): Updates the content generation goal
• setGuidelines(guidelines): Updates the guidelines array
• addGuideline(guideline): Adds a single guideline
• getGoal(): Returns the current goal
• getGuidelines(): Returns a copy of current guidelines

GeminiService

• generateContent(prompt, temperature?): Generates content with optional temperature control (default: 0.7)

Evaluator

• evaluateContent(content, goal, guidelines): Evaluates content against guidelines and returns score, feedback, and detailed results

How Evaluation Works

The application uses a sophisticated evaluation system:

1. AI-Powered Assessment: Uses Gemini AI to evaluate each guideline individually
2. Manual Score Calculation: Calculates the final score as (passed guidelines / total guidelines) * 100
3. Detailed Feedback: Provides specific feedback on which guidelines were met or missed
4. Smart Retry Logic: If content doesn't meet the threshold, retries with enhanced prompts that focus on previously failed guidelines
5. Transparent Results: Shows detailed breakdown of which guidelines passed/failed for each attempt