Cisco Agent Configuration Generator

A Retrieval-Augmented Generation (RAG) pipeline that ingests technical SQL + text specifications and generates configuration files or code snippets based on user instructions.

🏗️ Architecture

• Backend: Python 3.11 + Flask
• Database: MongoDB Atlas Cloud (vector search enabled)
• Vector Embeddings: sentence-transformers/all-MiniLM-L6-v2
• LLM Generation: Google Gemini (gemini-2.5-flash)
• Frontend: React single-page app

📋 Prerequisites

• Python 3.11+
• Node.js 16+ and npm
• MongoDB Atlas account with vector search enabled

🚀 Setup Instructions

1. Backend Setup

cd backend
pip install -r requirements.txt

2. Environment Configuration

Create a .env file in the root directory (already created with default values):

MONGODB_URI="mongodb+srv://<username>:<password>@cluster0.mongodb.net/?retryWrites=true&w=majority"
MONGODB_DB="ragdb"
MONGODB_COLLECTION="chunks"
EMBEDDING_MODEL="sentence-transformers/all-MiniLM-L6-v2"
GEMINI_API_KEY="your_gemini_api_key_here"
GEMINI_MODEL="gemini-2.5-flash"
GEN_TEMPERATURE="0.2"
TOP_K="5"

Note: Update MONGODB_URI with your actual MongoDB Atlas cluster hostname.

3. MongoDB Vector Index Setup

In MongoDB Atlas, create a vector search index on your collection:

1. Go to Atlas Search → Create Search Index
2. Use JSON Editor and paste:

{
  "fields": [
    {
      "type": "vector",
      "path": "embedding",
      "numDimensions": 384,
      "similarity": "cosine"
    }
  ]
}

3. Name the index: idx_embedding

4. Frontend Setup

cd frontend
npm install

5. Run the Application

Terminal 1 - Backend:

cd backend
python app.py

Terminal 2 - Frontend (development):

cd frontend
npm start

Or build for production:

cd frontend
npm run build

The backend will serve the frontend from the build folder when running in production mode.

📡 API Endpoints

GET /api/health

Health check endpoint to verify MongoDB connection.

Response:

{
  "status": "healthy",
  "mongodb": "connected",
  "database": "ragdb",
  "collection": "chunks"
}

POST /api/ingest

Upload and ingest SQL or text files.

Request: Multipart form data with file field (.sql or .txt)

Response:

{
  "status": "success",
  "message": "Ingested 5 chunks from example.sql",
  "chunks": 5,
  "source_id": "example.sql"
}

POST /api/generate-config

Generate configuration based on user instruction.

Request:

{
  "instruction": "Generate YAML config for firewall rule X"
}

Response:

{
  "config_output": "generated config code...",
  "references": [
    {
      "chunk_id": "...",
      "source_id": "example.sql",
      "text_preview": "...",
      "score": 0.95
    }
  ],
  "full_response": "..."
}

🎯 Usage

1. Upload Specifications: Use the upload section to upload .sql or .txt files containing technical specifications, commands, or expected configurations.

2. Generate Configurations: Enter an instruction in the textarea (e.g., "Generate YAML for command X") and click "Generate Config". The system will:

   • Embed your instruction
   • Retrieve the most relevant chunks from uploaded specs
   • Generate a configuration using the LLM
   • Display the result with context references

🔧 Features

• Intelligent Chunking: Automatically chunks SQL files by statements or text files by ~1000 characters
• Vector Search: Uses MongoDB Atlas vector search for semantic retrieval
• Context-Aware Generation: LLM generates configurations based on retrieved context
• Reference Tracking: Shows which chunks were used for generation
• Fallback Mechanism: Manual cosine similarity if vector index is not available

⚠️ Notes

• The first run will download the embedding model (~90MB)
• Ensure you have a valid GEMINI_API_KEY
• MongoDB Atlas vector search requires a dedicated cluster with vector search enabled

🐛 Troubleshooting

• MongoDB Connection Issues: Verify your connection string and network access in Atlas
• Vector Search Not Working: Check that the vector index idx_embedding is created
• Model Loading Errors: Ensure sufficient disk space and RAM
• CORS Errors: Make sure Flask-CORS is installed and configured