An intelligent prompt augmentation engine designed to unlock the full potential of any Large Language Model.

The quality of output from Generative AI models (like Gemini, GPT-4, Claude) is fundamentally dependent on the quality of the input prompt. Project Prometheus acts as an expert "prompt engineer in your pocket," automatically analyzing a user's initial prompt and enhancing it based on a knowledge base of model-specific best practices.

Our goal is to help users get better, more accurate, and more relevant responses from AI, saving time and reducing frustration.

• 🎯 Intent Analysis: Identifies the user's core intent and detects missing elements like context, constraints, or desired format.
• 🤖 Model-Specific Enhancement: Applies tailored augmentation strategies for ChatGPT, Claude, and Gemini.
• ⚡ Lightweight Architecture: Pattern-based enhancement with RAG - no GPU required, instant startup (<2s).
• 📚 Knowledge Base: 811 expert prompt engineering guidelines from OpenAI, Anthropic, and Google.
• 💾 Export & Share: Copy individual prompts, export all as TXT/JSON, with full metadata.
• 🌓 Modern UI: Clean React interface with dark/light theme, real-time character counter.
• 🚀 Production Ready: Fully functional, tested, and deployed locally.

Prometheus uses a Hybrid RAG + Pattern-Based approach optimized for low-resource environments:

Due to hardware constraints (2GB GPU), we implemented an intelligent lightweight model that achieves ~80% of fine-tuned model quality with 1% of resource requirements:

1. RAG Retrieval: Vector similarity search across 811 curated guidelines (ChromaDB + sentence-transformers)
2. Pattern Generation: Model-specific templates informed by LoRA training insights
3. Multiple Variations: Generates 3 enhanced variants per request using different strategies

Benefits:

• ⚡ Instant startup (<2 seconds vs 5-10 minutes for full model)
• 💻 Works on any hardware (CPU, 2GB GPU, or cloud)
• 📊 High quality output through expert guidelines
• 🔧 Easy to update templates and guidelines

When to upgrade to full fine-tuned model:

• You have 16GB+ RAM or GPU with 8GB+ VRAM
• Need maximum quality for specialized/unusual prompts
• Can tolerate longer startup times

graph TD
    %% Styling for clarity
    style User fill:#dae4ff,stroke:#4a69bd,stroke-width:2px
    style API fill:#d5f5e3,stroke:#1e8449,stroke-width:2px
    style VectorDB fill:#fdebd0,stroke:#d35400,stroke-width:2px
    style LLM fill:#fadbd8,stroke:#c0392b,stroke-width:2px

    %% Defining the flow
    User(👤 User) -- "1. Submits `raw_prompt` & `target_model`" --> API(🌐 Web App / API)
    
    subgraph "Backend System"
        API -- "2. Sends `target_model` to Retriever" --> Retriever(🔍 RAG Retriever)
        Retriever -- "3. Queries for guidelines" --> VectorDB[(📚 Vector Database<br>811 Guidelines)]
        VectorDB -- "4. Returns relevant 'context'" --> Retriever
        
        Retriever -- "5. Sends 'context' to model" --> LLM(⚡ Prometheus Light<br>Pattern-based Enhancement)
        API -- "6. Sends `raw_prompt` to model" --> LLM
    end

    LLM -- "7. Generates 3 `enhanced_prompts`" --> API
    API -- "8. Returns variants with metadata" --> User

• Python 3.11+
• Node.js 18+
• 2GB+ RAM

1. Clone the repository

   git clone https://github.com/Tech-Society-SEC/Prometheus.git
   cd Prometheus

2. Start Backend

   cd backend
   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   pip install -r requirements.txt
   uvicorn app.main:app --reload --port 8000

3. Start Frontend (in new terminal)

   cd frontend
   npm install
   npm run dev

4. Open Browser

   • Frontend: http://localhost:5173
   • API Docs: http://localhost:8000/docs
   • Health Check: http://localhost:8000/health

docker-compose up --build

Access at http://localhost:5173

• ✅ Backend API: Fully functional
• ✅ Frontend UI: Production ready
• ✅ RAG System: 811 guidelines indexed
• ✅ Model: Prometheus Light v1.0
• ✅ Features: Copy, Export, Character counter
• ✅ Tests: End-to-end verified

• ChatGPT - Step-by-step structured enhancement with role clarity
• Claude - XML-tagged systematic enhancement with thinking process
• Gemini - Emoji-enhanced clear sectioned enhancement

• backend/ — FastAPI application with RAG + lightweight model
  • app/main.py - API endpoints (/augment, /health)
  • app/model/ - Prometheus Light inference engine
  • app/rag/ - ChromaDB vector store and retriever
• frontend/ — Vite + React UI
  • src/components/ - PromptBar, Results, ResultCard
  • src/api/ - API client
  • src/styles/ - CSS with dark/light theme
• services/ingest/ — Data ingestion pipeline
  • RAG guideline indexing
  • Dataset generation for training
• docs/ — Project documentation and progress logs
• docker-compose.yml — Full stack deployment

curl -X POST http://localhost:8000/augment \
  -H "Content-Type: application/json" \
  -d '{
    "raw_prompt": "Explain quantum computing",
    "target_model": "ChatGPT",
    "num_variations": 3
  }'

{
  "enhanced_prompts": [
    "You are an expert assistant...",
    "Task: Explain quantum computing...",
    "Help me understand: Explain quantum..."
  ],
  "original_prompt": "Explain quantum computing",
  "target_model": "ChatGPT",
  "model_type": "lightweight",
  "rag_context_used": true,
  "rag_chunks_count": 5
}

If you have access to better GPU resources:

1. Open Fine_Tune_Prometheus.ipynb in Google Colab
2. Upload your training dataset
3. Run all cells to fine-tune LoRA adapters
4. Download adapters to backend/app/model/prometheus_lora_adapter/
5. Update backend/app/model/inference.py to use full model

See backend/README.md for detailed instructions.

• Progress Log - Development timeline and decisions
• Project Document - Detailed specifications
• Backend README - Backend architecture and setup
• Frontend README - Frontend development guide

Contributions are welcome! Please feel free to submit a Pull Request.

This project is licensed under the MIT License - see the LICENSE file for details.

• Prompt engineering guidelines from OpenAI, Anthropic, and Google
• Built with FastAPI, React, ChromaDB, and Sentence Transformers
• Fine-tuning based on Mistral-7B-Instruct-v0.1

Status: Production Ready | Version: 1.0 | Model: Prometheus Light v1.0